wehub-resource-sync
32 lines
919 B
C++
32 lines
919 B
C++
/*!
|
|
* Copyright (c) 2023-2025 by Contributors
|
|
* \file json_ffi/image_utils.h
|
|
* \brief The header of Image utils for JSON FFI Engine in MLC LLM.
|
|
*/
|
|
#ifndef MLC_LLM_JSON_FFI_IMAGE_UTILS_H_
|
|
#define MLC_LLM_JSON_FFI_IMAGE_UTILS_H_
|
|
|
|
#include <tvm/runtime/tensor.h>
|
|
|
|
#include <optional>
|
|
#include <string>
|
|
|
|
#include "../support/result.h"
|
|
|
|
namespace mlc {
|
|
namespace llm {
|
|
namespace json_ffi {
|
|
|
|
/*! \brief Load a base64 encoded image string into a CPU Tensor of shape {height, width, 3} */
|
|
Result<tvm::runtime::Tensor> LoadImageFromBase64(const std::string& base64_str);
|
|
|
|
/*! \brief Preprocess the CPU image for CLIP encoder and return an Tensor on the given device */
|
|
tvm::runtime::Tensor ClipPreprocessor(tvm::runtime::Tensor image_data, int target_size,
|
|
DLDevice device);
|
|
|
|
} // namespace json_ffi
|
|
} // namespace llm
|
|
} // namespace mlc
|
|
|
|
#endif // MLC_LLM_JSON_FFI_IMAGE_UTILS_H_
|