caffe卷积层代码注解

base_conv_layer.cpp

#include <algorithm>
#include <vector>

#include "caffe/filler.hpp"
#include "caffe/layer.hpp"
#include "caffe/util/im2col.hpp"
#include "caffe/util/math_functions.hpp"
#include "caffe/vision_layers.hpp"

namespace caffe {
// 执行流程是，首先执行LayerSetUp，然后执行Reshape
// 最后才是前传，反传



// 设置kernel的大小
// 设置stride
// 设置pad
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
      const vector<Blob<Dtype>*>& top) {
      
  // Configure the kernel size, padding, stride, and inputs.
  ConvolutionParameter conv_param = this->layer_param_.convolution_param();
  // 是否需要强制n维卷积
  force_nd_im2col_ = conv_param.force_nd_im2col();
  
  // 输入图像的通道是第几个轴
  channel_axis_ = bottom[0]->CanonicalAxisIndex(conv_param.axis());
  // 一般来说channel之后是图像的高度和宽度，这里称之为空间轴
  const int first_spatial_axis = channel_axis_ + 1;
  
  // 一共有几个轴，比如图像的blob是[b,c,h,w]那么num_axes=4
  // channel_axis = 1, first_spatial_axis=2
  const int num_axes = bottom[0]->num_axes();

  // 空间轴的个数
  // num_spatial_axes_=4-2=2
  num_spatial_axes_ = num_axes - first_spatial_axis;

  // num_spatial_axes_ >= 0
  CHECK_GE(num_spatial_axes_, 0);

  // 变量有shape的都是存放的是形状数据
  // 输入的blob的形状是[b,c,h]所以bottom_dim_blob_shape的维度是3
  vector<int> bottom_dim_blob_shape(1, num_spatial_axes_ + 1);
  
  // 空间的blob的形状是[h,w]，所以是max(num_spatial_axes_,1)=2
  vector<int> spatial_dim_blob_shape(1, std::max(num_spatial_axes_, 1));

  
  // Setup filter kernel dimensions (kernel_shape_).
  // filter kernel的
  // kernel的形状
  kernel_shape_.Reshape(spatial_dim_blob_shape);
  int* kernel_shape_data = kernel_shape_.mutable_cpu_data();
  if (conv_param.has_kernel_h() || conv_param.has_kernel_w()) {
    CHECK_EQ(num_spatial_axes_, 2)
        << "kernel_h & kernel_w can only be used for 2D convolution.";
    CHECK_EQ(0, conv_param.kernel_size_size())
        << "Either kernel_size or kernel_h/w should be specified; not both.";
    kernel_shape_data[0] = conv_param.kernel_h();
    kernel_shape_data[1] = conv_param.kernel_w();
  } else {// 正方形的kernel
    // kernel_size中的形状参数的个数，这里num_kernel_dims = 1
    const int num_kernel_dims = conv_param.kernel_size_size();
    CHECK(num_kernel_dims == 1 || num_kernel_dims == num_spatial_axes_)
        << "kernel_size must be specified once, or once per spatial dimension "
        << "(kernel_size specified " << num_kernel_dims << " times; "
        << num_spatial_axes_ << " spatial dims);";
	  // kernel_shape_data[0]和[1]=conv_param.kernel_size(0)
      for (int i = 0; i < num_spatial_axes_; ++i) {
        kernel_shape_data[i] =
            conv_param.kernel_size((num_kernel_dims == 1) ? 0 : i);
      }
  }

  // 检查kernel_shape_data是不是正确设置，即kernel的边长不能为<0的情况
  for (int i = 0; i < num_spatial_axes_; ++i) {
    CHECK_GT(kernel_shape_data[i], 0) << "Filter dimensions must be nonzero.";
  }

  
  // Setup stride dimensions (stride_).
  // 设置步长(stride)
  stride_.Reshape(spatial_dim_blob_shape);
  int* stride_data = stride_.mutable_cpu_data();
  // 是否是矩形的方框
  if (conv_param.has_stride_h() || conv_param.has_stride_w()) {
    CHECK_EQ(num_spatial_axes_, 2)
        << "stride_h & stride_w can only be used for 2D convolution.";
    CHECK_EQ(0, conv_param.stride_size())
        << "Either stride or stride_h/w should be specified; not both.";
    stride_data[0] = conv_param.stride_h();
    stride_data[1] = conv_param.stride_w();
  } else {// 正方形的方框
    // num_stride_dims = 1
    const int num_stride_dims = conv_param.stride_size();
    CHECK(num_stride_dims == 0 || num_stride_dims == 1 ||
          num_stride_dims == num_spatial_axes_)
        << "stride must be specified once, or once per spatial dimension "
        << "(stride specified " << num_stride_dims << " times; "
        << num_spatial_axes_ << " spatial dims);";
    const int kDefaultStride = 1;
    for (int i = 0; i < num_spatial_axes_; ++i) {
	  // stride_data[0]和[1] = conv_param.stride(0)
      stride_data[i] = (num_stride_dims == 0) ? kDefaultStride :
          conv_param.stride((num_stride_dims == 1) ? 0 : i);
      CHECK_GT(stride_data[i], 0) << "Stride dimensions must be nonzero.";
    }
  }

  
  // Setup pad dimensions (pad_).
  // 设置pad的数据
  pad_.Reshape(spatial_dim_blob_shape);
  int* pad_data = pad_.mutable_cpu_data();
  if (conv_param.has_pad_h() || conv_param.has_pad_w()) {
    CHECK_EQ(num_spatial_axes_, 2)
        << "pad_h & pad_w can only be used for 2D convolution.";
    CHECK_EQ(0, conv_param.pad_size())
        << "Either pad or pad_h/w should be specified; not both.";
    pad_data[0] = conv_param.pad_h();
    pad_data[1] = conv_param.pad_w();
  } else {
    const int num_pad_dims = conv_param.pad_size();
    CHECK(num_pad_dims == 0 || num_pad_dims == 1 ||
          num_pad_dims == num_spatial_axes_)
        << "pad must be specified once, or once per spatial dimension "
        << "(pad specified " << num_pad_dims << " times; "
        << num_spatial_axes_ << " spatial dims);";
    const int kDefaultPad = 0;
    for (int i = 0; i < num_spatial_axes_; ++i) {
	  // pad_data[0]和[1] = conv_param.pad[0]
      pad_data[i] = (num_pad_dims == 0) ? kDefaultPad :
          conv_param.pad((num_pad_dims == 1) ? 0 : i);
    }
  }
  // Special case: im2col is the identity for 1x1 convolution with stride 1
  // and no padding, so flag for skipping the buffer and transformation.
  // 这里是1x1卷积，kernel_shape_data[0]与[1]=1并且stride_data[0]与[1]=1
  // 且pad_data[0]与[1]=1
  is_1x1_ = true;
  for (int i = 0; i < num_spatial_axes_; ++i) {
    is_1x1_ &=
        kernel_shape_data[i] == 1 && stride_data[i] == 1 && pad_data[i] == 0;
    if (!is_1x1_) { break; }
  }

  
  // Configure output channels and groups.
  // 设置输出的通道以及卷积组的大小

  // 输入图像的通道数
  channels_ = bottom[0]->shape(channel_axis_);

  // 经过卷积之后的输出图像的通道数
  num_output_ = this->layer_param_.convolution_param().num_output();
  CHECK_GT(num_output_, 0);

  // 卷积组的大小
  group_ = this->layer_param_.convolution_param().group();

  // 检查合法性
  CHECK_EQ(channels_ % group_, 0);
  CHECK_EQ(num_output_ % group_, 0)
      << "Number of output should be multiples of group.";

  // 将输入通道和输出通道赋值给
  // conv_in_channels_和conv_out_channels_
  if (reverse_dimensions()) {// 是否要翻转维度，将输出作为输入，输入作为输出
    conv_out_channels_ = channels_;
    conv_in_channels_ = num_output_;
  } else {// 输入的channel和输出的channel
    conv_out_channels_ = num_output_;
    conv_in_channels_ = channels_;
  }


  // 设置权重的形状(包含了所有kernel)
  // Handle the parameters: weights and biases.
  // - blobs_[0] holds the filter weights
  // - blobs_[1] holds the biases (optional)

  // 权重形状 = [conv_out_channels_, conv_in_channels_/group, weight_h, weight_w]
  vector<int> weight_shape(2);
  weight_shape[0] = conv_out_channels_;
  weight_shape[1] = conv_in_channels_ / group_;// 将输入的channel分组
  for (int i = 0; i < num_spatial_axes_; ++i) { // num_spatial_axes_=2
    weight_shape.push_back(kernel_shape_data[i]);
  }

  // 偏置的形状 = [num_output_]
  // bias_term = 1
  bias_term_ = this->layer_param_.convolution_param().bias_term();
  vector<int> bias_shape(bias_term_, num_output_);
  if (this->blobs_.size() > 0) {
    CHECK_EQ(1 + bias_term_, this->blobs_.size())
        << "Incorrect number of weight blobs.";
    if (weight_shape != this->blobs_[0]->shape()) {
      Blob<Dtype> weight_shaped_blob(weight_shape);
      LOG(FATAL) << "Incorrect weight shape: expected shape "
          << weight_shaped_blob.shape_string() << "; instead, shape was "
          << this->blobs_[0]->shape_string();
    }
    if (bias_term_ && bias_shape != this->blobs_[1]->shape()) {
      Blob<Dtype> bias_shaped_blob(bias_shape);
      LOG(FATAL) << "Incorrect bias shape: expected shape "
          << bias_shaped_blob.shape_string() << "; instead, shape was "
          << this->blobs_[1]->shape_string();
    }
    LOG(INFO) << "Skipping parameter initialization";
  } else {
    if (bias_term_) {
	  // 如果有则blobs_[0]表示权重和[1]表示偏置
      this->blobs_.resize(2);
    } else {
      this->blobs_.resize(1);
    }

	  // 初始化权重
    // Initialize and fill the weights:
    // 权重的形状如下
    // output channels x input channels per-group x kernel height x kernel width
    this->blobs_[0].reset(new Blob<Dtype>(weight_shape));
    shared_ptr<Filler<Dtype> > weight_filler(GetFiller<Dtype>(
        this->layer_param_.convolution_param().weight_filler()));
    weight_filler->Fill(this->blobs_[0].get());
    // If necessary, initialize and fill the biases.
    // 初始化偏置
    if (bias_term_) {
      this->blobs_[1].reset(new Blob<Dtype>(bias_shape));
      shared_ptr<Filler<Dtype> > bias_filler(GetFiller<Dtype>(
          this->layer_param_.convolution_param().bias_filler()));
      bias_filler->Fill(this->blobs_[1].get());
    }
  }

  // count(1)表示从第2个维度到最后一个维度之间的数据的个数
  // 即kernel_dim_ = (conv_in_channels_/group)*weight_h*weight_w
  kernel_dim_ = this->blobs_[0]->count(1);

  // 计算卷积分组用的offset（所有kernel矩阵的元素总和）
  weight_offset_ = conv_out_channels_ * kernel_dim_ / group_;

  // Propagate gradients to the parameters (as directed by backward pass).
  this->param_propagate_down_.resize(this->blobs_.size(), true);
}


// 该函数干了一下几件事情
// 改变top的形状
// 设置卷积分组用的offset
// 设置经过im2col变换之后存放数据col_buffer_的形状
// bias_multiplier_将偏置扩展成矩阵
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
      const vector<Blob<Dtype>*>& top) {
  // 第一个空间轴是第几个
  const int first_spatial_axis = channel_axis_ + 1;
  CHECK_EQ(bottom[0]->num_axes(), first_spatial_axis + num_spatial_axes_)
      << "bottom num_axes may not change.";
  
  // num_ = batchsize
  num_ = bottom[0]->count(0, channel_axis_);
  CHECK_EQ(bottom[0]->shape(channel_axis_), channels_)
      << "Input size incompatible with convolution kernel.";

  
  // TODO: generalize to handle inputs of different shapes.
  for (int bottom_id = 1; bottom_id < bottom.size(); ++bottom_id) {
    CHECK(bottom[0]->shape() == bottom[bottom_id]->shape())
        << "All inputs must have the same shape.";
  }

  
  // Shape the tops.
  bottom_shape_ = &bottom[0]->shape();
  
  // 计算得到output_shape_
  // 即输出的图像的高度和宽度
  compute_output_shape();

  // 设置输出top的形状
  // 将batchsize压入到top_shape(就是channel之前的部分，也就是batchsize了)
  vector<int> top_shape(bottom[0]->shape().begin(),
      bottom[0]->shape().begin() + channel_axis_);
  // 输出的channel压入到top_shape
  top_shape.push_back(num_output_);
  // 压入输出图像的长度和宽度
  for (int i = 0; i < num_spatial_axes_; ++i) {
    top_shape.push_back(output_shape_[i]);
  }
  // 改变top的形状
  // batchsize X conv_out_channel X out_height X out_width
  for (int top_id = 0; top_id < top.size(); ++top_id) {
    top[top_id]->Reshape(top_shape);
  }

  // 反卷积reverse_dimensions才=1
  if (reverse_dimensions()) {
  	// conv_out_spatial_dim_= 输入图像的长度和宽度
    conv_out_spatial_dim_ = bottom[0]->count(first_spatial_axis);
  } else {
    // conv_out_spatial_dim_ = 卷积之后输出图像的长度和宽度
    conv_out_spatial_dim_ = top[0]->count(first_spatial_axis);
  }


  // 设置卷积分组用的offset
  // kernel_dim_ = (conv_in_channels_/group)*weight_h*weight_w
  col_offset_ = kernel_dim_ * conv_out_spatial_dim_;
  output_offset_ = conv_out_channels_ * conv_out_spatial_dim_ / group_;

  
  // Setup input dimensions (conv_input_shape_).
  vector<int> bottom_dim_blob_shape(1, num_spatial_axes_ + 1);
  // 卷积的输入的形状是三维的，因此num_spatial_axes_+1=3
  conv_input_shape_.Reshape(bottom_dim_blob_shape);
  int* conv_input_shape_data = conv_input_shape_.mutable_cpu_data();
  for (int i = 0; i < num_spatial_axes_ + 1; ++i) {
    if (reverse_dimensions()) {
	  // conv_input_shape_data = [卷积之后输出图像通道数, 卷积之后输出图像h, 卷积之后输出图像图像w]
      conv_input_shape_data[i] = top[0]->shape(channel_axis_ + i);
    } else {
      // conv_input_shape_data = [输入图像通道数, 输入图像h, 输入图像w]
      conv_input_shape_data[i] = bottom[0]->shape(channel_axis_ + i);
    }
  }

  
  // The im2col result buffer will only hold one image at a time to avoid
  // overly large memory usage. In the special case of 1x1 convolution
  // it goes lazily unused to save memory.
  col_buffer_shape_.clear();
  col_buffer_shape_.push_back(kernel_dim_ * group_);
  for (int i = 0; i < num_spatial_axes_; ++i) {
    if (reverse_dimensions()) {
	  // col_buffer_shape_ = [kernel_dim_, conv_in_spatial_dim_]
      col_buffer_shape_.push_back(input_shape(i + 1));
    } else {
      // col_buffer_shape_ = [kernel_dim_, conv_out_spatial_dim_]
      col_buffer_shape_.push_back(output_shape_[i]);
    }
  }

  // 经过im2col变换之后存放数据col_buffer_的形状
  col_buffer_.Reshape(col_buffer_shape_);

  // 输入的大小 bottom_dim_ = conv_in_channel X image_height X image_width
  bottom_dim_ = bottom[0]->count(channel_axis_);
  // 输出的大小 top_dim_ = conv_out_channel X out_height X out_width
  top_dim_ = top[0]->count(channel_axis_);


  // conv_in_channels_ = 输入图像的通道数
  // conv_out_spatial_dim_ = 卷积之后输出图像的长度和宽度
  num_kernels_im2col_ = conv_in_channels_ * conv_out_spatial_dim_;

  
  // top_dim_ = conv_out_channel x out_height x out_width
  // bottom_dim_ = conv_in_channel x image_height x image_width
  num_kernels_col2im_ = reverse_dimensions() ? top_dim_ : bottom_dim_;

  
  // Set up the all ones "bias multiplier" for adding biases by BLAS
  // bias_multiplier_将偏置扩展成矩阵
  // bias_multiplier_的形状是  1 x out_spatial_dim_
  out_spatial_dim_ = top[0]->count(first_spatial_axis);
  if (bias_term_) {
    vector<int> bias_multiplier_shape(1, out_spatial_dim_);
    bias_multiplier_.Reshape(bias_multiplier_shape);
    caffe_set(bias_multiplier_.count(), Dtype(1),
        bias_multiplier_.mutable_cpu_data());
  }
}



// 对batch中的一个进行卷积运算
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::forward_cpu_gemm(const Dtype* input,
    const Dtype* weights, Dtype* output, bool skip_im2col) {
  const Dtype* col_buff = input;
  // kernel_dim_ = input channels per-group x kernel height x kernel width
  
  if (!is_1x1_) {
    if (!skip_im2col) {
	  // 如果没有1x1卷积，也没有skip_im2col
	  // 则使用conv_im2col_cpu对使用卷积核滑动过程中的每一个kernel大小的图像块
	  // 变成一个列向量，形成一个
	  // height = kernel_dim_
	  // width = 卷积后图像heght*卷积后图像width
	  // 的矩阵
      conv_im2col_cpu(input, col_buffer_.mutable_cpu_data());
    }
    col_buff = col_buffer_.cpu_data();
  }

  // 使用caffe的cpu_gemm来进行计算
  for (int g = 0; g < group_; ++g) {
  	// 分组分别进行计算
  	// conv_out_channels_ / group_是每个卷积组的输出的channel
  	
  	// output[output_offset_ * g] = weights[weight_offset_ * g] X col_buff[col_offset_ * g]
  	// weights的形状是 [conv_out_channel x kernel_dim_]
  	// col_buff的形状是[kernel_dim_ x (卷积后图像高度乘以卷积后图像宽度)]
  	// output的形状自然就是conv_out_channel X (卷积后图像高度乘以卷积后图像宽度)
    caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, conv_out_channels_ /
        group_, conv_out_spatial_dim_, kernel_dim_,
        (Dtype)1., weights + weight_offset_ * g, col_buff + col_offset_ * g,
        (Dtype)0., output + output_offset_ * g);
  }
}

// 加上偏置
// output是经过卷积后的图像
// output = output + bias * bias_multiplier_
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::forward_cpu_bias(Dtype* output,
    const Dtype* bias) {
  // output = output + bias * bias_multiplier_
  // num_output 与 conv_out_channel的值相等
  // bias会转换为形状:num_output_ x 1
  // bias_multiplier_的形状是:1 x out_spatial_dim_
  // 那么这两货相乘之后的形状是:num_output_ x out_spatial_dim_
  caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, num_output_,
      out_spatial_dim_, 1, (Dtype)1., bias, bias_multiplier_.cpu_data(),
      (Dtype)1., output);
}


// 卷积的反传
// 卷积本来就是先用im2col转换为矩阵，然后执行的实际上就是y=W*x
// 对y关于x求偏导得到W
// 那么反传到下一层的梯度就是col_buff = W*output
// 再对col_buff用col2im转换为原图像即可完成反传
// 参数output就是top_diff, weight就是当前层的权重， input是bottom_diff
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::backward_cpu_gemm(const Dtype* output,
    const Dtype* weights, Dtype* input) {
  Dtype* col_buff = col_buffer_.mutable_cpu_data();
  if (is_1x1_) {
    col_buff = input;
  }
  for (int g = 0; g < group_; ++g) {
    // 计算的是col_buff[col_offset_ * g] = 
    // weights[weight_offset_ * g] x output[output_offset_ * g]
    caffe_cpu_gemm<Dtype>(CblasTrans, CblasNoTrans, kernel_dim_,
        conv_out_spatial_dim_, conv_out_channels_ / group_,
        (Dtype)1., weights + weight_offset_ * g, output + output_offset_ * g,
        (Dtype)0., col_buff + col_offset_ * g);
  }
  if (!is_1x1_) {
  	// 再用col2im转换回来
    conv_col2im_cpu(col_buff, input);
  }
}

// 这个是计算权重的更新
// 卷积实际上就是y=W*x+b
// 那么y关于W求偏导就是x
// 那么权重的更新就是weights += output*col_buff
// 参数input是bottom_data, output就是top_diff, weight就是当前层的权重 
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::weight_cpu_gemm(const Dtype* input,
    const Dtype* output, Dtype* weights) {
  const Dtype* col_buff = input;
  if (!is_1x1_) {
    conv_im2col_cpu(input, col_buffer_.mutable_cpu_data());
    col_buff = col_buffer_.cpu_data();
  }
  for (int g = 0; g < group_; ++g) {
  	// 计算的是weights[weight_offset_ * g] += 
  	// output[output_offset_ * g] X col_buff[col_offset_ * g]
    caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasTrans, conv_out_channels_ / group_,
        kernel_dim_, conv_out_spatial_dim_,
        (Dtype)1., output + output_offset_ * g, col_buff + col_offset_ * g,
        (Dtype)1., weights + weight_offset_ * g);
  }
}

// 偏置的反传
// 因为是y=Wx+b
// y关于b求偏导即为1
// 那么就是bias += input*bias_multiplier_
template <typename Dtype>
void BaseConvolutionLayer<Dtype>::backward_cpu_bias(Dtype* bias,
    const Dtype* input) {
  // input的形状:num_output_ * out_spatial_dim_
  // bias_multiplier_的形状:out_spatial_dim_ * 1
  // bias = input*bias_multiplier_+ bias
  caffe_cpu_gemv<Dtype>(CblasNoTrans, num_output_, out_spatial_dim_, 1.,
      input, bias_multiplier_.cpu_data(), 1., bias);
}

#ifndef CPU_ONLY

template <typename Dtype>
void BaseConvolutionLayer<Dtype>::forward_gpu_gemm(const Dtype* input,
    const Dtype* weights, Dtype* output, bool skip_im2col) {
  const Dtype* col_buff = input;
  if (!is_1x1_) {
    if (!skip_im2col) {
      conv_im2col_gpu(input, col_buffer_.mutable_gpu_data());
    }
    col_buff = col_buffer_.gpu_data();
  }
  for (int g = 0; g < group_; ++g) {
    caffe_gpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, conv_out_channels_ /
        group_, conv_out_spatial_dim_, kernel_dim_,
        (Dtype)1., weights + weight_offset_ * g, col_buff + col_offset_ * g,
        (Dtype)0., output + output_offset_ * g);
  }
}

template <typename Dtype>
void BaseConvolutionLayer<Dtype>::forward_gpu_bias(Dtype* output,
    const Dtype* bias) {
  caffe_gpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, num_output_,
      out_spatial_dim_, 1, (Dtype)1., bias, bias_multiplier_.gpu_data(),
      (Dtype)1., output);
}

template <typename Dtype>
void BaseConvolutionLayer<Dtype>::backward_gpu_gemm(const Dtype* output,
    const Dtype* weights, Dtype* input) {
  Dtype* col_buff = col_buffer_.mutable_gpu_data();
  if (is_1x1_) {
    col_buff = input;
  }
  for (int g = 0; g < group_; ++g) {
    caffe_gpu_gemm<Dtype>(CblasTrans, CblasNoTrans, kernel_dim_,
        conv_out_spatial_dim_, conv_out_channels_ / group_,
        (Dtype)1., weights + weight_offset_ * g, output + output_offset_ * g,
        (Dtype)0., col_buff + col_offset_ * g);
  }
  if (!is_1x1_) {
    conv_col2im_gpu(col_buff, input);
  }
}

template <typename Dtype>
void BaseConvolutionLayer<Dtype>::weight_gpu_gemm(const Dtype* input,
    const Dtype* output, Dtype* weights) {
  const Dtype* col_buff = input;
  if (!is_1x1_) {
    conv_im2col_gpu(input, col_buffer_.mutable_gpu_data());
    col_buff = col_buffer_.gpu_data();
  }
  for (int g = 0; g < group_; ++g) {
    caffe_gpu_gemm<Dtype>(CblasNoTrans, CblasTrans, conv_out_channels_ / group_,
        kernel_dim_, conv_out_spatial_dim_,
        (Dtype)1., output + output_offset_ * g, col_buff + col_offset_ * g,
        (Dtype)1., weights + weight_offset_ * g);
  }
}

template <typename Dtype>
void BaseConvolutionLayer<Dtype>::backward_gpu_bias(Dtype* bias,
    const Dtype* input) {
  caffe_gpu_gemv<Dtype>(CblasNoTrans, num_output_, out_spatial_dim_, 1.,
      input, bias_multiplier_.gpu_data(), 1., bias);
}

#endif  // !CPU_ONLY

INSTANTIATE_CLASS(BaseConvolutionLayer);

}  // namespace caffe

conv_layer.cpp

#include <vector>

#include "caffe/filler.hpp"
#include "caffe/layer.hpp"
#include "caffe/util/im2col.hpp"
#include "caffe/util/math_functions.hpp"
#include "caffe/vision_layers.hpp"

namespace caffe {

// Layer类首先会SetUp，然后会Reshape
// 基类在Reshape的时候会调用此函数
template <typename Dtype>
void ConvolutionLayer<Dtype>::compute_output_shape() {
  const int* kernel_shape_data = this->kernel_shape_.cpu_data();
  const int* stride_data = this->stride_.cpu_data();
  const int* pad_data = this->pad_.cpu_data();
  this->output_shape_.clear();
  for (int i = 0; i < this->num_spatial_axes_; ++i) {
    // i + 1 to skip channel axis
    // input_shape(i) = (*bottom_shape_)[channel_axis_ + i]
    const int input_dim = this->input_shape(i + 1);
    const int output_dim = (input_dim + 2 * pad_data[i] - kernel_shape_data[i])
        / stride_data[i] + 1;
    this->output_shape_.push_back(output_dim);
  }
}

// 卷积的前传
template <typename Dtype>
void ConvolutionLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
      const vector<Blob<Dtype>*>& top) {
  // 卷积核，你也可以理解为权重
  const Dtype* weight = this->blobs_[0]->cpu_data();

  // 每一个输入图像进行处理
  for (int i = 0; i < bottom.size(); ++i) {
    const Dtype* bottom_data = bottom[i]->cpu_data();
    Dtype* top_data = top[i]->mutable_cpu_data();
	// num_ = batchsize
    for (int n = 0; n < this->num_; ++n) {
	  // 基类的forward_cpu_gemm函数
	  // 计算的是top_data[n * this->top_dim_] = 
	  // weights X bottom_data[n * this->bottom_dim_]
      this->forward_cpu_gemm(bottom_data + n * this->bottom_dim_, weight,
          top_data + n * this->top_dim_);
      if (this->bias_term_) {
        const Dtype* bias = this->blobs_[1]->cpu_data();
        this->forward_cpu_bias(top_data + n * this->top_dim_, bias);
      }
    }
  }
}

template <typename Dtype>
void ConvolutionLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
      const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom) {
  const Dtype* weight = this->blobs_[0]->cpu_data();
  Dtype* weight_diff = this->blobs_[0]->mutable_cpu_diff();
  for (int i = 0; i < top.size(); ++i) {
    const Dtype* top_diff = top[i]->cpu_diff();
    const Dtype* bottom_data = bottom[i]->cpu_data();
    Dtype* bottom_diff = bottom[i]->mutable_cpu_diff();
    // Bias gradient, if necessary.
    // 反传偏置
    if (this->bias_term_ && this->param_propagate_down_[1]) {
      Dtype* bias_diff = this->blobs_[1]->mutable_cpu_diff();
	  // num_ = batchsize
      for (int n = 0; n < this->num_; ++n) {
        this->backward_cpu_bias(bias_diff, top_diff + n * this->top_dim_);//由于一个batch的所有样本的梯度是累加的，
      }                                                                   //所以bias_diff会在一个batch的反向传播中不断累加,是否会求平均值？
    }

	// 要判断是否反传
    if (this->param_propagate_down_[0] || propagate_down[i]) {
		// num_ = batchsize
      for (int n = 0; n < this->num_; ++n) {
        // gradient w.r.t. weight. Note that we will accumulate diffs.
        if (this->param_propagate_down_[0]) {
		  // 计算权重更新
          this->weight_cpu_gemm(bottom_data + n * this->bottom_dim_,  //由于一个batch的所有样本的梯度是累加的，
              top_diff + n * this->top_dim_, weight_diff);             //所以weight_diff会在一个batch的反向传播中不断累加,是否会求平均值？
        }
        // gradient w.r.t. bottom data, if necessary.
        // 计算关于卷积层输入的梯度，放入bottom_diff，便于反传到前一层
        if (propagate_down[i]) {
          this->backward_cpu_gemm(top_diff + n * this->top_dim_, weight,    
              bottom_diff + n * this->bottom_dim_);                            
        }
      }
    }
  }
}

#ifdef CPU_ONLY
STUB_GPU(ConvolutionLayer);
#endif

INSTANTIATE_CLASS(ConvolutionLayer);

}  // namespace caffe

举例