microsoft/onnxruntime-extensions
Publicmirrored fromhttps://github.com/microsoft/onnxruntime-extensionsAvailable
operators/cv2/super_resolution_preprocess.cc
91lines · modecode
| 1 | // Copyright (c) Microsoft Corporation. All rights reserved. |
| 2 | // Licensed under the MIT License. |
| 3 | |
| 4 | #include "super_resolution_preprocess.hpp" |
| 5 | #include "string_utils.h" |
| 6 | |
| 7 | #include <opencv2/core.hpp> |
| 8 | #include <opencv2/imgproc.hpp> |
| 9 | #include <opencv2/imgcodecs.hpp> |
| 10 | |
| 11 | #include <cstdint> |
| 12 | |
| 13 | KernelSuperResolutionPreProcess::KernelSuperResolutionPreProcess(const OrtApi& api) : BaseKernel(api) {} |
| 14 | |
| 15 | void KernelSuperResolutionPreProcess::Compute(OrtKernelContext* context) { |
| 16 | // Setup inputs |
| 17 | const OrtValue* const inputs = ort_.KernelContext_GetInput(context, 0ULL); |
| 18 | OrtTensorDimensions dimensions(ort_, inputs); |
| 19 | if (dimensions.size() != 1ULL) { |
| 20 | throw std::runtime_error("Only raw image formats are supported."); |
| 21 | } |
| 22 | |
| 23 | // Get data & the length |
| 24 | const uint8_t* const encoded_bgr_image_data = ort_.GetTensorData<uint8_t>(inputs); |
| 25 | |
| 26 | OrtTensorTypeAndShapeInfo* const input_info = ort_.GetTensorTypeAndShape(inputs); |
| 27 | const int64_t encoded_bgr_image_data_len = ort_.GetTensorShapeElementCount(input_info); |
| 28 | ort_.ReleaseTensorTypeAndShapeInfo(input_info); |
| 29 | |
| 30 | // Decode the image |
| 31 | const std::vector<int32_t> encoded_bgr_image_sizes{1, static_cast<int32_t>(encoded_bgr_image_data_len)}; |
| 32 | const cv::Mat encoded_bgr_image(encoded_bgr_image_sizes, CV_8UC1, |
| 33 | const_cast<void*>(static_cast<const void*>(encoded_bgr_image_data))); |
| 34 | // OpenCV decodes images in BGR format. |
| 35 | // Ref: https://stackoverflow.com/a/44359400 |
| 36 | const cv::Mat decoded_bgr_image = cv::imdecode(encoded_bgr_image, cv::IMREAD_COLOR); |
| 37 | |
| 38 | cv::Mat normalized_bgr_image; |
| 39 | decoded_bgr_image.convertTo(normalized_bgr_image, CV_32F); |
| 40 | |
| 41 | cv::Mat ycrcb_image; |
| 42 | cv::cvtColor(normalized_bgr_image, ycrcb_image, cv::COLOR_BGR2YCrCb); |
| 43 | |
| 44 | cv::Mat channels[3]; |
| 45 | cv::split(ycrcb_image, channels); |
| 46 | channels[0] /= 255.0; |
| 47 | |
| 48 | // Setup output & copy to destination |
| 49 | for (int32_t i = 0; i < 3; ++i) { |
| 50 | const cv::Mat& channel = channels[i]; |
| 51 | const cv::Size size = channel.size(); |
| 52 | |
| 53 | const std::vector<int64_t> output_dimensions{1LL, 1LL, size.height, size.width}; |
| 54 | OrtValue* const output_value = ort_.KernelContext_GetOutput( |
| 55 | context, i, output_dimensions.data(), output_dimensions.size()); |
| 56 | float* const data = ort_.GetTensorMutableData<float>(output_value); |
| 57 | memcpy(data, channel.data, channel.total() * channel.elemSize()); |
| 58 | } |
| 59 | } |
| 60 | |
| 61 | const char* CustomOpSuperResolutionPreProcess::GetName() const { |
| 62 | return "SuperResolutionPreProcess"; |
| 63 | } |
| 64 | |
| 65 | size_t CustomOpSuperResolutionPreProcess::GetInputTypeCount() const { |
| 66 | return 1; |
| 67 | } |
| 68 | |
| 69 | ONNXTensorElementDataType CustomOpSuperResolutionPreProcess::GetInputType(size_t index) const { |
| 70 | switch (index) { |
| 71 | case 0: |
| 72 | return ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8; |
| 73 | default: |
| 74 | ORTX_CXX_API_THROW(MakeString("Unexpected input index ", index), ORT_INVALID_ARGUMENT); |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | size_t CustomOpSuperResolutionPreProcess::GetOutputTypeCount() const { |
| 79 | return 3; |
| 80 | } |
| 81 | |
| 82 | ONNXTensorElementDataType CustomOpSuperResolutionPreProcess::GetOutputType(size_t index) const { |
| 83 | switch (index) { |
| 84 | case 0: |
| 85 | case 1: |
| 86 | case 2: |
| 87 | return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT; |
| 88 | default: |
| 89 | ORTX_CXX_API_THROW(MakeString("Unexpected output index ", index), ORT_INVALID_ARGUMENT); |
| 90 | } |
| 91 | } |
| 92 | |