chore: import upstream snapshot with attribution

plugin/detectionLayerPlugin/CMakeLists.txt

@@ -0,0 +1,22 @@
+#
+# SPDX-FileCopyrightText: Copyright (c) 1993-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+add_plugin_source(
+    detectionLayerPlugin.cpp
+    detectionLayerPlugin.h
+)
+

plugin/detectionLayerPlugin/DetectionLayer_PluginConfig.yaml

@@ -0,0 +1,55 @@
+#
+# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+---
+name: DetectionLayer_TRT
+interface: "IPluginV2Ext"
+versions:
+  "1":
+    attributes:
+      - num_classes
+      - keep_topk
+      - score_threshold
+      - iou_threshold
+    attribute_types:
+      num_classes: int32
+      keep_topk: int32
+      score_threshold: float32
+      iou_threshold: float32
+    attribute_length:
+      num_classes: 1
+      keep_topk: 1
+      score_threshold: 1
+      iou_threshold: 1
+    attribute_options:
+      num_classes:
+        min: "0"
+        max: "=pinf"
+      keep_topk:
+        min: "0"
+        max: "=pinf"
+      score_threshold:
+        min: "=0"
+        max: "=pinf"
+      iou_threshold:
+        min: "0"
+        max: "=pinf"
+    attributes_required:
+      - num_classes
+      - keep_topk
+      - score_threshold
+      - iou_threshold
+...

plugin/detectionLayerPlugin/README.md

@@ -0,0 +1,78 @@
+# DetectionLayer Plugin [DEPRECATED]
+
+**This plugin is deprecated since TensorRT 10.12 and will be removed in a future release. No alternatives are planned to be provided.**
+
+**Table Of Contents**
+- [Description](#description)
+    * [Structure](#structure)
+- [Parameters](#parameters)
+- [Additional resources](#additional-resources)
+- [License](#license)
+- [Changelog](#changelog)
+- [Known issues](#known-issues)
+
+## Description
+
+The `DetectionLayer` plugin performs bounding boxes refinement of MaskRCNN's detection head and generate the final detection output of MaskRCNN. It is used in sampleMaskRCNN.  
+
+
+### Structure
+
+This plugin supports the NCHW format. It takes three input tensors: `delta_bbox`, `score` and `roi`
+
+`delta_bbox` is the refinement information of roi boxes generated from `ProposalLayer`. `delta_bbox` tensor's shape is `[N, rois, num_classes*4, 1, 1]` where `N` is batch size,
+`rois` is the total number of ROI boxes candidates per image, and `num_classes*4` means 4 refinement elements (`[dy, dx, dh, dw]`) for each roi box as different classes.
+
+`score` is the predicted class scores of ROI boxes generated from `ProposalLayer` of shape `[N, rois, num_classes, 1, 1]`. There is `argmax`operation in `Detectionlayer` to determine the final class of detection
+candidates.   
+
+`roi` is the coordinates of ROI boxes candidates from `ProposalLayer` of shape `[N, rois, 4]`. 
+
+This plugin generates output of shape `[N, keep_topk, 6]` where `keep_topk` is the maximum number of detections left after NMS and '6' means 6 elements of an detection `[y1, x1, y2, x2,
+class_label, score]`
+
+## Parameters
+
+This plugin has the plugin creator class `DetectionlayerPluginCreator` and the plugin class `Detectionlayer`.
+  
+The following parameters were used to create `Detectionlayer` instance:
+
+| Type               | Parameter                          | Description
+|--------------------|------------------------------------|--------------------------------------------------------
+|`int`               |`num_classes`                       |Number of detection classes(including `background`). `num_classes=81` for COCO dataset
+|`int`               |`keep_topk`                         |Number of detections will be kept after NMS.  
+|`float`             |`score_threshold`                   |Confidence threshold value. This plugin will drop a detection if its class confidence(score) is under "score_threshold". 
+|`float`             |`iou_threshold`                     |IOU threshold value used in NMS.
+
+## Limitations
+
+The number of anchors is capped at 1024 to support embedded devices with smaller shared memory capacity.
+
+To enable support for a device with higher memory, calls to `sortPerClass`, `PerClassNMS` and `KeepTopKGather` can be modified in `RefineBatchClassNMS` ([maskRCNNKernels.cu](https://github.com/NVIDIA/TensorRT/blob/main/plugin/common/kernels/maskRCNNKernels.cu)).
+
+## Additional resources
+
+The following resources provide a deeper understanding of the `Detectionlayer` plugin:
+
+- [MaskRCNN](https://github.com/matterport/Mask_RCNN)
+
+
+## License
+
+For terms and conditions for use, reproduction, and distribution, see the [TensorRT Software License Agreement](https://docs.nvidia.com/deeplearning/sdk/tensorrt-sla/index.html) 
+documentation.
+
+
+## Changelog
+
+May 2025
+Add deprecation note.
+
+January 2022: The [Limitations](#limitations) section was added to this `README.md` file to document limitations of the plugin related to the maximum number of anchors it can support.  
+
+June 2019: First release of this `README.md` file.
+
+
+## Known issues
+
+There are no known issues in this plugin.

plugin/detectionLayerPlugin/detectionLayerPlugin.cpp

@@ -0,0 +1,356 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 1993-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "detectionLayerPlugin.h"
+#include "common/plugin.h"
+
+#include <memory>
+#include <string_view>
+
+using namespace nvinfer1;
+using namespace plugin;
+using nvinfer1::plugin::DetectionLayer;
+using nvinfer1::plugin::DetectionLayerPluginCreator;
+
+namespace
+{
+char const* const kDETECTIONLAYER_PLUGIN_VERSION{"1"};
+char const* const kDETECTIONLAYER_PLUGIN_NAME{"DetectionLayer_TRT"};
+} // namespace
+
+DetectionLayerPluginCreator::DetectionLayerPluginCreator()
+{
+    mPluginAttributes.clear();
+    mPluginAttributes.emplace_back(PluginField("num_classes", nullptr, PluginFieldType::kINT32, 1));
+    mPluginAttributes.emplace_back(PluginField("keep_topk", nullptr, PluginFieldType::kINT32, 1));
+    mPluginAttributes.emplace_back(PluginField("score_threshold", nullptr, PluginFieldType::kFLOAT32, 1));
+    mPluginAttributes.emplace_back(PluginField("iou_threshold", nullptr, PluginFieldType::kFLOAT32, 1));
+
+    mFC.nbFields = mPluginAttributes.size();
+    mFC.fields = mPluginAttributes.data();
+}
+
+char const* DetectionLayerPluginCreator::getPluginName() const noexcept
+{
+    return kDETECTIONLAYER_PLUGIN_NAME;
+}
+
+char const* DetectionLayerPluginCreator::getPluginVersion() const noexcept
+{
+    return kDETECTIONLAYER_PLUGIN_VERSION;
+}
+
+PluginFieldCollection const* DetectionLayerPluginCreator::getFieldNames() noexcept
+{
+    return &mFC;
+}
+
+IPluginV2Ext* DetectionLayerPluginCreator::createPlugin(char const* /*name*/, PluginFieldCollection const* fc) noexcept
+{
+    using namespace std::string_view_literals;
+    try
+    {
+        PLUGIN_VALIDATE(fc != nullptr);
+        plugin::validateRequiredAttributesExist({"num_classes", "keep_topk", "score_threshold", "iou_threshold"}, fc);
+        PluginField const* fields = fc->fields;
+        for (int32_t i = 0; i < fc->nbFields; ++i)
+        {
+            std::string_view const attrName = fields[i].name;
+            if (attrName == "num_classes"sv)
+            {
+                PLUGIN_VALIDATE(fields[i].type == PluginFieldType::kINT32);
+                mNbClasses = *(static_cast<int32_t const*>(fields[i].data));
+            }
+            if (attrName == "keep_topk"sv)
+            {
+                PLUGIN_VALIDATE(fields[i].type == PluginFieldType::kINT32);
+                mKeepTopK = *(static_cast<int32_t const*>(fields[i].data));
+            }
+            if (attrName == "score_threshold"sv)
+            {
+                PLUGIN_VALIDATE(fields[i].type == PluginFieldType::kFLOAT32);
+                mScoreThreshold = *(static_cast<float const*>(fields[i].data));
+            }
+            if (attrName == "iou_threshold"sv)
+            {
+                PLUGIN_VALIDATE(fields[i].type == PluginFieldType::kFLOAT32);
+                mIOUThreshold = *(static_cast<float const*>(fields[i].data));
+            }
+        }
+        return new DetectionLayer(mNbClasses, mKeepTopK, mScoreThreshold, mIOUThreshold);
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return nullptr;
+}
+
+IPluginV2Ext* DetectionLayerPluginCreator::deserializePlugin(
+    char const* /*name*/, void const* data, size_t length) noexcept
+{
+    try
+    {
+        return new DetectionLayer(data, length);
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return nullptr;
+}
+
+DetectionLayer::DetectionLayer(int32_t numClasses, int32_t keepTopk, float scoreThreshold, float iouThreshold)
+    : mNbClasses(numClasses)
+    , mKeepTopK(keepTopk)
+    , mScoreThreshold(scoreThreshold)
+    , mIOUThreshold(iouThreshold)
+{
+    mBackgroundLabel = 0;
+    PLUGIN_VALIDATE(mNbClasses > 0);
+    PLUGIN_VALIDATE(mKeepTopK > 0);
+    PLUGIN_VALIDATE(mScoreThreshold >= 0.F);
+    PLUGIN_VALIDATE(mIOUThreshold > 0.F);
+
+    mParam.backgroundLabelId = 0;
+    mParam.numClasses = mNbClasses;
+    mParam.keepTopK = mKeepTopK;
+    mParam.scoreThreshold = mScoreThreshold;
+    mParam.iouThreshold = mIOUThreshold;
+
+    mType = DataType::kFLOAT;
+}
+
+int32_t DetectionLayer::getNbOutputs() const noexcept
+{
+    return 1;
+}
+
+int32_t DetectionLayer::initialize() noexcept
+{
+    try
+    {
+        // Init the mValidCnt and mDecodedBboxes for max batch size.
+        std::vector<int32_t> tempValidCnt(mMaxBatchSize, mAnchorsCnt);
+
+        mValidCnt = std::make_shared<CudaBind<int32_t>>(mMaxBatchSize);
+
+        PLUGIN_CUASSERT(cudaMemcpy(mValidCnt->mPtr, static_cast<void*>(tempValidCnt.data()),
+            sizeof(int32_t) * mMaxBatchSize, cudaMemcpyHostToDevice));
+
+        return STATUS_SUCCESS;
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return STATUS_FAILURE;
+}
+
+void DetectionLayer::terminate() noexcept {}
+
+void DetectionLayer::destroy() noexcept
+{
+    delete this;
+}
+
+bool DetectionLayer::supportsFormat(DataType type, PluginFormat format) const noexcept
+{
+    return (type == DataType::kFLOAT && format == PluginFormat::kLINEAR);
+}
+
+char const* DetectionLayer::getPluginType() const noexcept
+{
+    return kDETECTIONLAYER_PLUGIN_NAME;
+}
+
+char const* DetectionLayer::getPluginVersion() const noexcept
+{
+    return kDETECTIONLAYER_PLUGIN_VERSION;
+}
+
+IPluginV2Ext* DetectionLayer::clone() const noexcept
+{
+    try
+    {
+        auto plugin = std::make_unique<DetectionLayer>(*this);
+        plugin->setPluginNamespace(mNameSpace.c_str());
+        return plugin.release();
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return nullptr;
+}
+
+void DetectionLayer::setPluginNamespace(char const* libNamespace) noexcept
+{
+    try
+    {
+        PLUGIN_VALIDATE(libNamespace != nullptr);
+        mNameSpace = libNamespace;
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+}
+
+char const* DetectionLayer::getPluginNamespace() const noexcept
+{
+    return mNameSpace.c_str();
+}
+
+size_t DetectionLayer::getSerializationSize() const noexcept
+{
+    return sizeof(int32_t) * 2 + sizeof(float) * 2 + sizeof(int32_t) * 2;
+}
+
+void DetectionLayer::serialize(void* buffer) const noexcept
+{
+    auto* d = reinterpret_cast<uint8_t*>(buffer);
+    auto* const a = d;
+    write(d, mNbClasses);
+    write(d, mKeepTopK);
+    write(d, mScoreThreshold);
+    write(d, mIOUThreshold);
+    write(d, mMaxBatchSize);
+    write(d, mAnchorsCnt);
+    PLUGIN_ASSERT(d == a + getSerializationSize());
+}
+
+DetectionLayer::DetectionLayer(void const* data, size_t length)
+{
+    auto const* d = reinterpret_cast<uint8_t const*>(data);
+    auto const* const a = d;
+    mNbClasses = read<int32_t>(d);
+    mKeepTopK = read<int32_t>(d);
+    mScoreThreshold = read<float>(d);
+    mIOUThreshold = read<float>(d);
+    mMaxBatchSize = read<int32_t>(d);
+    mAnchorsCnt = read<int32_t>(d);
+    PLUGIN_VALIDATE(d == a + length);
+
+    mParam.backgroundLabelId = 0;
+    mParam.numClasses = mNbClasses;
+    mParam.keepTopK = mKeepTopK;
+    mParam.scoreThreshold = mScoreThreshold;
+    mParam.iouThreshold = mIOUThreshold;
+
+    mType = DataType::kFLOAT;
+}
+
+void DetectionLayer::checkValidInputs(nvinfer1::Dims const* inputs, int32_t nbInputDims)
+{
+    // classifier_delta_bbox[N, anchors, num_classes*4, 1, 1]
+    // classifier_class[N, anchors, num_classes, 1, 1]
+    // rpn_rois[N, anchors, 4]
+    PLUGIN_VALIDATE(nbInputDims == 3);
+    // delta_bbox
+    PLUGIN_VALIDATE(inputs[0].nbDims == 4 && inputs[0].d[1] == mNbClasses * 4);
+    // score
+    PLUGIN_VALIDATE(inputs[1].nbDims == 4 && inputs[1].d[1] == mNbClasses);
+    // roi
+    PLUGIN_VALIDATE(inputs[2].nbDims == 2 && inputs[2].d[1] == 4);
+}
+
+size_t DetectionLayer::getWorkspaceSize(int32_t batchSize) const noexcept
+{
+    RefineDetectionWorkSpace refine(batchSize, mAnchorsCnt, mParam, mType);
+    return refine.totalSize;
+}
+
+Dims DetectionLayer::getOutputDimensions(int32_t index, Dims const* inputs, int32_t nbInputDims) noexcept
+{
+    try
+    {
+        checkValidInputs(inputs, nbInputDims);
+        PLUGIN_VALIDATE(index == 0);
+        // [N, anchors, (y1, x1, y2, x2, class_id, score)]
+        return {2, {mKeepTopK, 6}};
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return Dims{};
+}
+
+int32_t DetectionLayer::enqueue(
+    int32_t batchSize, void const* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept
+{
+    try
+    {
+        PLUGIN_VALIDATE(inputs != nullptr);
+        PLUGIN_VALIDATE(outputs != nullptr);
+        void* detections = outputs[0];
+
+        // refine detection
+        RefineDetectionWorkSpace refDetcWorkspace(batchSize, mAnchorsCnt, mParam, mType);
+        cudaError_t status = RefineBatchClassNMS(stream, batchSize, mAnchorsCnt,
+            DataType::kFLOAT, // mType,
+            mParam, refDetcWorkspace, workspace,
+            inputs[1],       // inputs[InScore]
+            inputs[0],       // inputs[InDelta],
+            mValidCnt->mPtr, // inputs[InCountValid],
+            inputs[2],       // inputs[ROI]
+            detections);
+
+        return status;
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+    return STATUS_FAILURE;
+}
+
+DataType DetectionLayer::getOutputDataType(
+    int32_t index, nvinfer1::DataType const* inputTypes, int32_t nbInputs) const noexcept
+{
+    // Only DataType::kFLOAT is acceptable by the plugin layer.
+    return DataType::kFLOAT;
+}
+
+// Configure the layer with input and output data types.
+void DetectionLayer::configurePlugin(Dims const* inputDims, int32_t nbInputs, Dims const* outputDims, int32_t nbOutputs,
+    DataType const* inputTypes, DataType const* outputTypes, bool const* inputIsBroadcast,
+    bool const* outputIsBroadcast, PluginFormat floatFormat, int32_t maxBatchSize) noexcept
+{
+    try
+    {
+        checkValidInputs(inputDims, nbInputs);
+        PLUGIN_VALIDATE(inputDims[0].d[0] == inputDims[1].d[0] && inputDims[1].d[0] == inputDims[2].d[0]);
+
+        mAnchorsCnt = inputDims[2].d[0];
+        mType = inputTypes[0];
+        mMaxBatchSize = maxBatchSize;
+    }
+    catch (std::exception const& e)
+    {
+        caughtError(e);
+    }
+}
+
+// Attach the plugin object to an execution context and grant the plugin the access to some context resource.
+void DetectionLayer::attachToContext(
+    cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator) noexcept
+{
+}
+
+// Detach the plugin object from its execution context.
+void DetectionLayer::detachFromContext() noexcept {}

plugin/detectionLayerPlugin/detectionLayerPlugin.h

@@ -0,0 +1,130 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 1993-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef TRT_DETECTION_LAYER_PLUGIN_H
+#define TRT_DETECTION_LAYER_PLUGIN_H
+#include <cuda_runtime_api.h>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "NvInfer.h"
+#include "NvInferPlugin.h"
+#include "common/kernels/maskRCNNKernels.h"
+
+namespace nvinfer1
+{
+namespace plugin
+{
+
+class DetectionLayer : public IPluginV2Ext
+{
+public:
+    DetectionLayer(int32_t numClasses, int32_t keepTopk, float scoreThreshold, float iouThreshold);
+
+    DetectionLayer(void const* data, size_t length);
+
+    ~DetectionLayer() override = default;
+
+    int32_t getNbOutputs() const noexcept override;
+
+    Dims getOutputDimensions(int32_t index, Dims const* inputs, int32_t nbInputDims) noexcept override;
+
+    int32_t initialize() noexcept override;
+
+    void terminate() noexcept override;
+
+    void destroy() noexcept override;
+
+    size_t getWorkspaceSize(int32_t maxBatchSize) const noexcept override;
+
+    int32_t enqueue(int32_t batchSize, void const* const* inputs, void* const* outputs, void* workspace,
+        cudaStream_t stream) noexcept override;
+
+    size_t getSerializationSize() const noexcept override;
+
+    void serialize(void* buffer) const noexcept override;
+
+    bool supportsFormat(DataType type, PluginFormat format) const noexcept override;
+
+    char const* getPluginType() const noexcept override;
+
+    char const* getPluginVersion() const noexcept override;
+
+    IPluginV2Ext* clone() const noexcept override;
+
+    void setPluginNamespace(char const* libNamespace) noexcept override;
+
+    char const* getPluginNamespace() const noexcept override;
+
+    DataType getOutputDataType(
+        int32_t index, nvinfer1::DataType const* inputTypes, int32_t nbInputs) const noexcept override;
+
+    void attachToContext(
+        cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator) noexcept override;
+
+    void configurePlugin(Dims const* inputDims, int32_t nbInputs, Dims const* outputDims, int32_t nbOutputs,
+        DataType const* inputTypes, DataType const* outputTypes, bool const* inputIsBroadcast,
+        bool const* outputIsBroadcast, PluginFormat floatFormat, int32_t maxBatchSize) noexcept override;
+
+    void detachFromContext() noexcept override;
+
+private:
+    void checkValidInputs(nvinfer1::Dims const* inputs, int32_t nbInputDims);
+
+    int32_t mBackgroundLabel;
+    int32_t mNbClasses;
+    int32_t mKeepTopK;
+    float mScoreThreshold;
+    float mIOUThreshold;
+
+    int32_t mMaxBatchSize;
+    int32_t mAnchorsCnt;
+    std::shared_ptr<CudaBind<int32_t>> mValidCnt; // valid cnt = number of input rois for every image.
+    nvinfer1::DataType mType;
+    RefineNMSParameters mParam;
+
+    std::string mNameSpace;
+};
+
+class DetectionLayerPluginCreator : public nvinfer1::pluginInternal::BaseCreator
+{
+public:
+    DetectionLayerPluginCreator();
+
+    ~DetectionLayerPluginCreator() override {}
+
+    char const* getPluginName() const noexcept override;
+
+    char const* getPluginVersion() const noexcept override;
+
+    PluginFieldCollection const* getFieldNames() noexcept override;
+
+    IPluginV2Ext* createPlugin(char const* name, PluginFieldCollection const* fc) noexcept override;
+
+    IPluginV2Ext* deserializePlugin(char const* name, void const* data, size_t length) noexcept override;
+
+private:
+    PluginFieldCollection mFC;
+    int32_t mNbClasses;
+    int32_t mKeepTopK;
+    float mScoreThreshold;
+    float mIOUThreshold;
+    std::vector<PluginField> mPluginAttributes;
+};
+} // namespace plugin
+} // namespace nvinfer1
+#endif // TRT_DETECTION_LAYER_PLUGIN_H