accuracy_layer

#include <functional>
#include <utility>
#include <vector>

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

namespace caffe {

template <typename Dtype>
void AccuracyLayer<Dtype>::LayerSetUp(
  const vector<Blob<Dtype>*>& bottom, const vector<Blob<Dtype>*>& top) {
  top_k_ = this->layer_param_.accuracy_param().top_k();//獲得k，也就是正确類别排前k名算個入acc

  has_ignore_label_ =
    this->layer_param_.accuracy_param().has_ignore_label();//有沒有要忽略的标簽
  if (has_ignore_label_) {
    ignore_label_ = this->layer_param_.accuracy_param().ignore_label();
  }
}
/*定義中關于axis的說明：
axis指出在預測blob中，哪一維是label軸，如(N x C x H x W)的blob，axis=0，則N為label對應的次元。
axis=1,則C為label對應的次元，而剩下的N為outer樣本數量， H x W為inner樣本數量。由代碼可知，
當axis=k時outer_num_=blob.shape[0,..,k)，inner_num_=blob.shape[k+1,..,shape.size)。一般的，
label blob的次元為(N x C)，N為樣本數量，C為标簽數量（即類别個數）。
axis=1,outer_num_=N,inner_num_=shape[2,2)=1(即沒有inner)
*/
template <typename Dtype>
void AccuracyLayer<Dtype>::Reshape(
  const vector<Blob<Dtype>*>& bottom, const vector<Blob<Dtype>*>& top) {
  CHECK_LE(top_k_, bottom[0]->count() / bottom[1]->count())//要取的k不能比總類别數大
      << "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的坐标軸
  outer_num_ = bottom[0]->count(0, label_axis_);//基本可以了解為batch中的樣本數
  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<int> top_shape(0);  // Accuracy is a scalar; 0 axes.
  top[0]->Reshape(top_shape);//top[0]是總體樣本正确率，标量top[1]為每個類别的正确率，向量
  if (top.size() > 1) {
    // Per-class accuracy is a vector; 1 axes.
    vector<int> 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 <typename Dtype>
void AccuracyLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
    const vector<Blob<Dtype>*>& top) {
  Dtype accuracy = 0;
  const Dtype* bottom_data = bottom[0]->cpu_data();//樣本數*标簽個數（也就是最後一個全連結的輸出層節點個數）
  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<Dtype> maxval(top_k_+1);
  vector<int> 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) {//對于每個樣本
    for (int j = 0; j < inner_num_; ++j) {
      const int label_value =
          static_cast<int>(bottom_label[i * inner_num_ + j]);//第i個樣本的label
      if (has_ignore_label_ && label_value == ignore_label_) {//如果這個類别被忽略就計算下一個。
        continue;
      }
      if (top.size() > 1) ++nums_buffer_.mutable_cpu_data()[label_value];//batch中每個類别的總樣本數，為了計算類内正确率
      DCHECK_GE(label_value, 0);
      DCHECK_LT(label_value, num_labels);
      // Top-k accuracy
      std::vector<std::pair<Dtype, int> > 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));//完成帶序号的排序
      }
      std::partial_sort(
          bottom_data_vector.begin(), bottom_data_vector.begin() + top_k_,
          bottom_data_vector.end(), std::greater<std::pair<Dtype, int> >());
      // 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在預測的前k個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中沒有某一類樣本就把這類樣本的正确率設定為0，不然的話就正常計算
          : 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