Caffe中im2col的实现解析

这里，我是将Caffe中im2col的解析过程直接拉了出来，使用C++进行了输出，方便理解。代码如下：

#include<iostream>

using namespace std;

bool is_a_ge_zero_and_a_lt_b(int a,int b)
{
    if(a>=0 && a <b) return true;
    return false;
}

void im2col_cpu(const float* data_im, const int channels,
    const int height, const int width, const int kernel_h, const int kernel_w,
    const int pad_h, const int pad_w,
    const int stride_h, const int stride_w,
    const int dilation_h, const int dilation_w,
    float* data_col) {
  const int output_h = (height + 2 * pad_h -
    (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1;
  const int output_w = (width + 2 * pad_w -
    (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1;
  const int channel_size = height * width;
  for (int channel = channels; channel--; data_im += channel_size) {
    for (int kernel_row = 0; kernel_row < kernel_h; kernel_row++) {
      for (int kernel_col = 0; kernel_col < kernel_w; kernel_col++) {
        int input_row = -pad_h + kernel_row * dilation_h;
        for (int output_rows = output_h; output_rows; output_rows--) {
          if (!is_a_ge_zero_and_a_lt_b(input_row, height)) {
            for (int output_cols = output_w; output_cols; output_cols--) {
              *(data_col++) = 0;
            }
          } else {
            int input_col = -pad_w + kernel_col * dilation_w;
            for (int output_col = output_w; output_col; output_col--) {
              if (is_a_ge_zero_and_a_lt_b(input_col, width)) {
                *(data_col++) = data_im[input_row * width + input_col];
              } else {
                *(data_col++) = 0;
              }
              input_col += stride_w;
            }
          }
          input_row += stride_h;
        }
      }
    }
  }
}


int main()
{
     float* data_im;
    int height=5;
    int width=5;   
    int kernel_h=3;   
    int kernel_w=3;
    int pad_h=1;   
    int pad_w=1;
    int stride_h=1;   
    int stride_w=1;
    int dilation_h=1;   
    int dilation_w=1;
    float* data_col;
    int channels =3;
    const int output_h = (height + 2 * pad_h -
    (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1;
      const int output_w = (width + 2 * pad_w -
    (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1;
    data_im = new float[channels*height*width];
    data_col = new float[channels*output_h*output_w*kernel_h*kernel_w];
    
    //init input image data
    for(int m=0;m<channels;++m)
    {
      for(int i=0;i<height;++i)
      {
        for(int j=0;j<width;++j)
        {
          data_im[m*width*height+i*width+j] = m*width*height+ i*width +j;
          cout <<data_im[m*width*height+i*width+j] <<' ';
        }
        cout <<endl;
      }
    }
    
    im2col_cpu(data_im, channels,
     height,width, kernel_h, kernel_w,
    pad_h, pad_w,
    stride_h, stride_w,
    dilation_h, dilation_w,
     data_col);
    cout <<channels<<endl;
    cout <<output_h<<endl;
    cout <<output_w<<endl;
    cout <<kernel_h<<endl;
    cout <<kernel_w<<endl;
   // cout <<"error"<<endl;
    for(int i=0;i<kernel_w*kernel_h*channels;++i)
    {    
        for(int j=0;j<output_w*output_h;++j)
        {
            cout <<data_col[i*output_w*output_h+j]<<' ';
        }
        cout <<endl;
    }

    return 0;
}

多通道卷积的图像别人已经给过很多了，大家可以搜到的基本都来自于一篇。这里附上一个我自己的理解过程，和程序的输出是完全一致的