fast rcnn 代码解析（一）

、

输入为一张图像和2000个候选框，conv5的特征和2000个候选框同时输入到Roi Pooling层中，将不同大小的框映射到conv5特征图，并统一为相同的尺寸，其实就是将大小不同的矩形框映射成大小相同的矩形框，Roi Pooling层如下所示：

由于2000个矩形框中的坐标对应的是原始图像，首先坐标映射到conv5的feature map上，就是把各个坐标除以输入图片与feature map的大小的比值，得到了feature map上的box坐标后，我们使用pooling得到输出；由于输入的图片大小不一，所以这里我们使用的spp pooling，spp pooling在pooling的过程中需要计算pooling后的结果对应的两个像素点反映社到feature map上所占的范围，然后在那个范围中进行取max或者取average。

在\caffe-fast-rcnn\src\caffe\layers\roi_pooling_layer.cu中查看具体实现过程如下

// LayerSetUp
template 
void ROIPoolingLayer::LayerSetUp(const vector*>& bottom,
      const vector*>& top) {
  ROIPoolingParameter roi_pool_param = this->layer_param_.roi_pooling_param();
  //经过Pooling后的feature map的高
  pooled_height_ = roi_pool_param.pooled_h();
  //经过Pooling后的feature map的宽
  pooled_width_ = roi_pool_param.pooled_w();
  //输入图片与feature map之前的比值，这个feature map指roi pooling层的输入
  spatial_scale_ = roi_pool_param.spatial_scale();
}
//Reshape
template 
void ROIPoolingLayer::Reshape(const vector*>& bottom,
      const vector*>& top) {
  //输入的feature map的channel数
  channels_ = bottom[0]->channels();
  //输入的feature map的高
  height_ = bottom[0]->height();
  //输入的feature map的宽
  width_ = bottom[0]->width();
  //设置输出的形状NCHW，N=ROI的个数，C=channels_，H=pooled_height_，W=pooled_width_
  top[0]->Reshape(bottom[1]->num(), channels_, pooled_height_,
      pooled_width_);
  //max_idx_的形状与top一致
  max_idx_.Reshape(bottom[1]->num(), channels_, pooled_height_,
      pooled_width_);
}
//Forward
template 
void ROIPoolingLayer::Forward_cpu(const vector*>& bottom,
      const vector*>& top) {
  //输入有两部分组成，data和rois
  const Dtype* bottom_data = bottom[0]->cpu_data();
  const Dtype* bottom_rois = bottom[1]->cpu_data();
  // Number of ROIs
  int num_rois = bottom[1]->num();
  int batch_size = bottom[0]->num();
  int top_count = top[0]->count();
  Dtype* top_data = top[0]->mutable_cpu_data();
  caffe_set(top_count, Dtype(-FLT_MAX), top_data);
  int* argmax_data = max_idx_.mutable_cpu_data();
  caffe_set(top_count, -1, argmax_data);

  // For each ROI R = [batch_index x1 y1 x2 y2]: max pool over R
  for (int n = 0; n < num_rois; ++n) {
    int roi_batch_ind = bottom_rois[0];
    //把原图的坐标映射到feature map上面
    int roi_start_w = round(bottom_rois[1] * spatial_scale_);
    int roi_start_h = round(bottom_rois[2] * spatial_scale_);
    int roi_end_w = round(bottom_rois[3] * spatial_scale_);
    int roi_end_h = round(bottom_rois[4] * spatial_scale_);
    //计算每个roi在feature map上面的大小
    int roi_height = max(roi_end_h - roi_start_h + 1, 1);
    int roi_width = max(roi_end_w - roi_start_w + 1, 1);
    //pooling之后的feature map的一个值对应于pooling之前的feature map上的大小
    //注：由于roi的大小不一致，所以每次都需要计算一次
    const Dtype bin_size_h = static_cast(roi_height)
                             / static_cast(pooled_height_);
    const Dtype bin_size_w = static_cast(roi_width)
                             / static_cast(pooled_width_);
    //找到对应的roi的feature map，如果input data的batch size为1
    //那么roi_batch_ind=0
    const Dtype* batch_data = bottom_data + bottom[0]->offset(roi_batch_ind);
    //pooling的过程是针对每一个channel的，所以需要循环遍历
    for (int c = 0; c < channels_; ++c) {
      //计算output的每一个值，所以需要遍历一遍output，然后求出所有值
      for (int ph = 0; ph < pooled_height_; ++ph) {
        for (int pw = 0; pw < pooled_width_; ++pw) {
          // Compute pooling region for this output unit:
          //  start (included) = floor(ph * roi_height / pooled_height_)
          //  end (excluded) = ceil((ph + 1) * roi_height / pooled_height_)
          // 计算output上的一点对应于input上面区域的大小[hstart, wstart, hend, wend]
          int hstart = static_cast(floor(static_cast(ph)
                                              * bin_size_h));
          int hend = static_cast(ceil(static_cast(ph + 1)
                                           * bin_size_h));
          int wstart = static_cast(floor(static_cast(pw)
                                              * bin_size_w));
          int wend = static_cast(ceil(static_cast(pw + 1)
                                           * bin_size_w));
          //将映射后的区域平动到对应的位置[hstart, wstart, hend, wend]
          hstart = min(max(hstart + roi_start_h, 0), height_);
          hend = min(max(hend + roi_start_h, 0), height_);
          wstart = min(max(wstart + roi_start_w, 0), width_);
          wend = min(max(wend + roi_start_w, 0), width_);
          //如果映射后的矩形框不符合
          bool is_empty = (hend <= hstart) || (wend <= wstart);
          //pool_index指的是此时计算的output的值对应于output的位置
          const int pool_index = ph * pooled_width_ + pw;
          //如果矩形不符合，此处output的值设为0，此处的对应于输入区域的最大值为-1
          if (is_empty) {
            top_data[pool_index] = 0;
            argmax_data[pool_index] = -1;
          }
          //遍历output的值对应于input的区域块
          for (int h = hstart; h < hend; ++h) {
            for (int w = wstart; w < wend; ++w) {
             // 对应于input上的位置
              const int index = h * width_ + w;
              //计算区域块的最大值，保存在output对应的位置上
              //同时记录最大值的索引
              if (batch_data[index] > top_data[pool_index]) {
                top_data[pool_index] = batch_data[index];
                argmax_data[pool_index] = index;
              }
            }
          }
        }
      }
      // Increment all data pointers by one channel
      batch_data += bottom[0]->offset(0, 1);
      top_data += top[0]->offset(0, 1);
      argmax_data += max_idx_.offset(0, 1);
    }
    // Increment ROI data pointer
    bottom_rois += bottom[1]->offset(1);
  }
}

如上图所示，输入层一共有5个输出，该层是一个python层，源代码在fast-rcnn-master\lib\roi_data_layer\layer.py中：

    def setup(self, bottom, top):
        """Setup the RoIDataLayer."""

        # parse the layer parameter string, which must be valid YAML
        # 将五个输出，reshape为对应的大小 
        layer_params = yaml.load(self.param_str_)

        self._num_classes = layer_params['num_classes']

        self._name_to_top_map = {
            'data': 0,
            'rois': 1,
            'labels': 2}

        # data blob: holds a batch of N images, each with 3 channels
        # The height and width (100 x 100) are dummy values
        top[0].reshape(1, 3, 100, 100)

        # rois blob: holds R regions of interest, each is a 5-tuple
        # (n, x1, y1, x2, y2) specifying an image batch index n and a
        # rectangle (x1, y1, x2, y2)
        top[1].reshape(1, 5)

        # labels blob: R categorical labels in [0, ..., K] for K foreground
        # classes plus background
        top[2].reshape(1)

        if cfg.TRAIN.BBOX_REG:
            self._name_to_top_map['bbox_targets'] = 3
            self._name_to_top_map['bbox_loss_weights'] = 4

            # bbox_targets blob: R bounding-box regression targets with 4
            # targets per class
            top[3].reshape(1, self._num_classes * 4)

            # bbox_loss_weights blob: At most 4 targets per roi are active;
            # thisbinary vector sepcifies the subset of active targets
            top[4].reshape(1, self._num_classes * 4)

    def forward(self, bottom, top):
        """Get blobs and copy them into this layer's top blob vector."""
        # 进入下一个minibatch 
        blobs = self._get_next_minibatch()

        for blob_name, blob in blobs.iteritems():
            top_ind = self._name_to_top_map[blob_name]
            # Reshape net's input blobs
            top[top_ind].reshape(*(blob.shape))
            # Copy data into net's input blobs
            top[top_ind].data[...] = blob.astype(np.float32, copy=False)

    def backward(self, top, propagate_down, bottom):
        """This layer does not propagate gradients."""
        pass

    def reshape(self, bottom, top):

        pass

  

联合的损失函数，分类层为互熵损失，回归层为平滑的L1损失

两个输出层，一个对每个RoI输出离散概率分布，类别k=21：

一个输出bounding box回归的位移：
k表示类别的索引，前两个参数是指相对于object proposal尺度不变的平移，后两个参数是指对数空间中相对于object proposal的高与宽。把这两个输出的损失写到一起：

k*是真实类别，式中第一项是分类损失，第二项是定位损失，L由R个输出取均值而来.对于分类loss，是一个N+1路的softmax输出，其中的N是类别个数，1是背景。对于回归loss，是一个4xN路输出的regressor，也就是说对于每个类别都会训练一个单独的regressor，这里regressor的loss不是L2的，而是一个平滑的L1，形式如下：

参考博客：

http://blog.csdn.net/lanran2/article/details/60143861

http://blog.csdn.net/langb2014/article/details/52794804