FastDeploy/fastdeploy/vision/detection/contrib/yolov5lite.h

// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// 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.

#pragma once
#include "fastdeploy/fastdeploy_model.h"
#include "fastdeploy/vision/common/processors/transform.h"
#include "fastdeploy/vision/common/result.h"

namespace fastdeploy {
namespace vision {
namespace detection {

class FASTDEPLOY_DECL YOLOv5Lite : public FastDeployModel {
 public:
  YOLOv5Lite(const std::string& model_file, const std::string& params_file = "",
             const RuntimeOption& custom_option = RuntimeOption(),
             const ModelFormat& model_format = ModelFormat::ONNX);

  ~YOLOv5Lite();

  virtual std::string ModelName() const { return "YOLOv5-Lite"; }

  virtual bool Predict(cv::Mat* im, DetectionResult* result,
                       float conf_threshold = 0.45,
                       float nms_iou_threshold = 0.25);

  void UseCudaPreprocessing(int max_img_size = 3840 * 2160);

  // tuple of (width, height)
  std::vector<int> size;
  // padding value, size should be same with Channels
  std::vector<float> padding_value;
  // only pad to the minimum rectange which height and width is times of stride
  bool is_mini_pad;
  // while is_mini_pad = false and is_no_pad = true, will resize the image to
  // the set size
  bool is_no_pad;
  // if is_scale_up is false, the input image only can be zoom out, the maximum
  // resize scale cannot exceed 1.0
  bool is_scale_up;
  // padding stride, for is_mini_pad
  int stride;
  // for offseting the boxes by classes when using NMS
  float max_wh;
  // downsample strides for YOLOv5Lite to generate anchors, will take
  // (8,16,32) as default values, might have stride=64.
  std::vector<int> downsample_strides;
  // anchors parameters, downsample_strides will take
  // (8,16,32), each stride has three anchors with width and hight.
  std::vector<std::vector<float>> anchor_config;
  // whether the model_file was exported with decode module. The official
  // YOLOv5Lite/export.py script will export ONNX file without
  // decode module. Please set it 'true' manually if the model file
  // was exported with decode module.
  // false : ONNX files without decode module.
  // true : ONNX file with decode module.
  bool is_decode_exported;

 private:
  // necessary parameters for GenerateAnchors to generate anchors when ONNX file
  // without decode module.
  struct Anchor {
    int grid0;
    int grid1;
    int stride;
    float anchor_w;
    float anchor_h;
  };

  bool Initialize();

  bool Preprocess(Mat* mat, FDTensor* output,
                  std::map<std::string, std::array<float, 2>>* im_info);

  bool CudaPreprocess(Mat* mat, FDTensor* output,
                      std::map<std::string, std::array<float, 2>>* im_info);

  bool Postprocess(FDTensor& infer_result, DetectionResult* result,
                   const std::map<std::string, std::array<float, 2>>& im_info,
                   float conf_threshold, float nms_iou_threshold);

  // the official YOLOv5Lite/export.py will export ONNX file without decode
  // module.
  // this fuction support the postporocess for ONNX file without decode module.
  // set the `is_decode_exported = false`, this function will work.
  bool PostprocessWithDecode(
      FDTensor& infer_result, DetectionResult* result,
      const std::map<std::string, std::array<float, 2>>& im_info,
      float conf_threshold, float nms_iou_threshold);

  void LetterBox(Mat* mat, const std::vector<int>& size,
                 const std::vector<float>& color, bool _auto,
                 bool scale_fill = false, bool scale_up = true,
                 int stride = 32);

  // generate anchors for decodeing when ONNX file without decode module.
  void GenerateAnchors(const std::vector<int>& size,
                       const std::vector<int>& downsample_strides,
                       std::vector<Anchor>* anchors, const int num_anchors = 3);

  // whether to inference with dynamic shape (e.g ONNX export with dynamic shape
  // or not.)
  // while is_dynamic_shape if 'false', is_mini_pad will force 'false'. This
  // value will
  // auto check by fastdeploy after the internal Runtime already initialized.
  bool is_dynamic_input_;
  // CUDA host buffer for input image
  uint8_t* input_img_cuda_buffer_host_ = nullptr;
  // CUDA device buffer for input image
  uint8_t* input_img_cuda_buffer_device_ = nullptr;
  // CUDA device buffer for TRT input tensor
  float* input_tensor_cuda_buffer_device_ = nullptr;
  // Whether to use CUDA preprocessing
  bool use_cuda_preprocessing_ = false;
};
}  // namespace detection
}  // namespace vision
}  // namespace fastdeploy