语义分割学习笔记（五）——DenseImageData 数据层

1 参数

message DenseImageDataParameter {
  // Specify the data source file.
  optional string source = 1;
  // Specify the batch size.
  optional uint32 batch_size = 2;
  // The rand_skip variable is for the data layer to skip a few data points
  // to avoid all asynchronous sgd clients to start at the same point. The skip
  // point would be set as rand_skip * rand(0,1). Note that rand_skip should not
  // be larger than the number of keys in the database.
  optional uint32 rand_skip = 3 [default = 0];
  // Whether or not ImageLayer should shuffle the list of files at every epoch.
  optional bool shuffle = 4 [default = false];
  // It will also resize images if new_height or new_width are not zero.
  optional uint32 new_height = 5 [default = 0];
  optional uint32 new_width = 6 [default = 0];
  // Label gets divided by this factor, to train the encoder architecture of ENet.
  optional uint32 label_divide_factor = 13 [default = 1];
  // Specify if the images are color or gray
  optional bool is_color = 7 [default = true];
  optional string mean_file = 8;
  optional string root_folder = 9 [default = ""];
  optional bool mirror = 10 [default = false];
  optional uint32 crop_width = 11 [default = 0];
  optional uint32 crop_height = 12 [default = 0];
}

参数：
  label_divide_factor对标签图采样比例，SegNet没有label_divide_factor参数，ENet版本caffe有该参数

2 label_divide_factor对应源码层次修改

 //直接设定读图大小
    cv::Mat cv_lab = ReadImageToCVMat(root_folder + lines_[lines_id_].second,
            new_height/label_divide_factor, new_width/label_divide_factor, false, true);
			
	//设定top[1]大小
	top[1]->Reshape(batch_size, 1, height/label_divide_factor, width/label_divide_factor);
    this->prefetch_label_.Reshape(batch_size, 1, height/label_divide_factor, width/label_divide_factor);
    this->transformed_label_.Reshape(1, 1, height/label_divide_factor, width/label_divide_factor);

3 ReadImageToCVMat 源码在io.hpp  io.cpp中实现：

	#ifdef USE_OPENCV
	cv::Mat ReadImageToCVMat(const string& filename,
		const int height, const int width, const bool is_color,
		const bool nearest_neighbour_interp) {
	  cv::Mat cv_img;
	  int cv_read_flag = (is_color ? CV_LOAD_IMAGE_COLOR :
		CV_LOAD_IMAGE_GRAYSCALE);
	  cv::Mat cv_img_origin = cv::imread(filename, cv_read_flag);
	  if (!cv_img_origin.data) {
		LOG(ERROR) << "Could not open or find file " << filename;
		return cv_img_origin;
	  }
	  if (height > 0 && width > 0) {
		int cv_interp_flag = nearest_neighbour_interp ? CV_INTER_NN :
														CV_INTER_LINEAR;
		cv::resize(cv_img_origin, cv_img, cv::Size(width, height), 0, 0,
			cv_interp_flag);
	  } else {
		cv_img = cv_img_origin;
	  }
	  return cv_img;
	}

	cv::Mat ReadImageToCVMat(const string& filename,
		const int height, const int width, const bool is_color) {
	  return ReadImageToCVMat(filename, height, width, is_color, false);
	}

	cv::Mat ReadImageToCVMat(const string& filename,
		const int height, const int width) {
	  return ReadImageToCVMat(filename, height, width, true);
	}

	cv::Mat ReadImageToCVMat(const string& filename,
		const bool is_color) {
	  return ReadImageToCVMat(filename, 0, 0, is_color);
	}

	cv::Mat ReadImageToCVMat(const string& filename) {
	  return ReadImageToCVMat(filename, 0, 0, true);
	}