accuracy_layer

#include 
#include 
#include 

#include "caffe/layers/accuracy_layer.hpp"
#include "caffe/util/math_functions.hpp"

namespace caffe {

template 
void AccuracyLayer::LayerSetUp(
  const vector*>& bottom, const vector*>& top) {
  top_k_ = this->layer_param_.accuracy_param().top_k();//获得k，也就是正确类别排前k名算个入acc

  has_ignore_label_ =
    this->layer_param_.accuracy_param().has_ignore_label();//有没有要忽略的标签
  if (has_ignore_label_) {
    ignore_label_ = this->layer_param_.accuracy_param().ignore_label();
  }
}
/*定义中关于axis的说明：
axis指出在预测blob中，哪一维是label轴，如(N x C x H x W)的blob，axis=0，则N为label对应的维度。
axis=1,则C为label对应的维度，而剩下的N为outer样本数量， H x W为inner样本数量。由代码可知，
当axis=k时outer_num_=blob.shape[0,..,k)，inner_num_=blob.shape[k+1,..,shape.size)。一般的，
label blob的维度为(N x C)，N为样本数量，C为标签数量（即类别个数）。
axis=1,outer_num_=N,inner_num_=shape[2,2)=1(即没有inner)
*/
template 
void AccuracyLayer::Reshape(
  const vector*>& bottom, const vector*>& top) {
  CHECK_LE(top_k_, bottom[0]->count() / bottom[1]->count())//要取的k不能比总类别数大
      << "top_k must be less than or equal to the number of classes.";
  label_axis_ =
      bottom[0]->CanonicalAxisIndex(this->layer_param_.accuracy_param().axis());//label的坐标轴
  outer_num_ = bottom[0]->count(0, label_axis_);//基本可以理解为batch中的样本数
  inner_num_ = bottom[0]->count(label_axis_ + 1);//1
  CHECK_EQ(outer_num_ * inner_num_, bottom[1]->count())
      << "Number of labels must match number of predictions; "
      << "e.g., if label axis == 1 and prediction shape is (N, C, H, W), "
      << "label count (number of labels) must be N*H*W, "
      << "with integer values in {0, 1, ..., C-1}.";
  vector top_shape(0);  // Accuracy is a scalar; 0 axes.
  top[0]->Reshape(top_shape);//top[0]是总体样本正确率，标量top[1]为每个类别的正确率，向量
  if (top.size() > 1) {
    // Per-class accuracy is a vector; 1 axes.
    vector top_shape_per_class(1);
    top_shape_per_class[0] = bottom[0]->shape(label_axis_);
    top[1]->Reshape(top_shape_per_class);
    nums_buffer_.Reshape(top_shape_per_class);
  }
}

template 
void AccuracyLayer::Forward_cpu(const vector*>& bottom,
    const vector*>& top) {
  Dtype accuracy = 0;
  const Dtype* bottom_data = bottom[0]->cpu_data();//样本数*标签个数（也就是最后一个全链接的输出层节点个数）
  const Dtype* bottom_label = bottom[1]->cpu_data();
  const int dim = bottom[0]->count() / outer_num_;
  const int num_labels = bottom[0]->shape(label_axis_);
  vector maxval(top_k_+1);
  vector max_id(top_k_+1);
  if (top.size() > 1) {
    caffe_set(nums_buffer_.count(), Dtype(0), nums_buffer_.mutable_cpu_data());
    caffe_set(top[1]->count(), Dtype(0), top[1]->mutable_cpu_data());
  }
  int count = 0;
  for (int i = 0; i < outer_num_; ++i) {//对于每个样本
    for (int j = 0; j < inner_num_; ++j) {
      const int label_value =
          static_cast(bottom_label[i * inner_num_ + j]);//第i个样本的label
      if (has_ignore_label_ && label_value == ignore_label_) {//如果这个类别被忽略就计算下一个。
        continue;
      }
      if (top.size() > 1) ++nums_buffer_.mutable_cpu_data()[label_value];//batch中每个类别的总样本数，为了计算类内正确率
      DCHECK_GE(label_value, 0);
      DCHECK_LT(label_value, num_labels);
      // Top-k accuracy
      std::vector > bottom_data_vector;
      for (int k = 0; k < num_labels; ++k) {
        bottom_data_vector.push_back(std::make_pair(
            bottom_data[i * dim + k * inner_num_ + j], k));//完成带序号的排序
      }
      std::partial_sort(
          bottom_data_vector.begin(), bottom_data_vector.begin() + top_k_,
          bottom_data_vector.end(), std::greater >());
      // check if true label is in top k predictions
      for (int k = 0; k < top_k_; k++) {
        if (bottom_data_vector[k].second == label_value) { //如果标定的label在预测的前k个label中
          ++accuracy;
          if (top.size() > 1) ++top[1]->mutable_cpu_data()[label_value];
          break;
        }
      }
      ++count;
    }
  }

  // LOG(INFO) << "Accuracy: " << accuracy;
  top[0]->mutable_cpu_data()[0] = accuracy / count;
  if (top.size() > 1) {
    for (int i = 0; i < top[1]->count(); ++i) {
      top[1]->mutable_cpu_data()[i] =
          nums_buffer_.cpu_data()[i] == 0 ? 0 //batch中没有某一类样本就把这类样本的正确率设置为0，不然的话就正常计算
          : top[1]->cpu_data()[i] / nums_buffer_.cpu_data()[i];
    }
  }
  // Accuracy layer should not be used as a loss function.
}

INSTANTIATE_CLASS(AccuracyLayer);
REGISTER_LAYER_CLASS(Accuracy);

}  // namespace caffe