Caffe源码解读: Softmax_loss_Layer的前向与反向传播

最新推荐文章于 2024-08-06 14:07:22 发布

faithenXX

最新推荐文章于 2024-08-06 14:07:22 发布

阅读量884

点赞数

分类专栏： caffe

本文链接：https://blog.csdn.net/zyf19930610/article/details/71210074

版权

caffe 专栏收录该内容

14 篇文章 0 订阅

订阅专栏

1,前向传播

分为两步：1，计算softmax概率prob_data，直接使用softmaxlayer的forward函数；

2，计算loss，采用交叉熵，即每个第i类数据的loss为-log(prob(i))。

template <typename Dtype>
void SoftmaxWithLossLayer<Dtype>::Forward_cpu(
    const vector<Blob<Dtype>*>& bottom, const vector<Blob<Dtype>*>& top) {
  // The forward pass computes the softmax prob values.
  softmax_layer_->Forward(softmax_bottom_vec_, softmax_top_vec_);//直接使用softmax_layer->forward()
  const Dtype* prob_data = prob_.cpu_data();                     //概率数据
  const Dtype* label = bottom[1]->cpu_data();                    //真实标签
  int dim = prob_.count() / outer_num_;
  int count = 0;
  Dtype loss = 0;
  for (int i = 0; i < outer_num_; ++i) {
    for (int j = 0; j < inner_num_; j++) {
      const int label_value = static_cast<int>(label[i * inner_num_ + j]);
      if (has_ignore_label_ && label_value == ignore_label_) {
        continue;
      }
      DCHECK_GE(label_value, 0);
      DCHECK_LT(label_value, prob_.shape(softmax_axis_));
      loss -= log(std::max(prob_data[i * dim + label_value * inner_num_ + j],
                           Dtype(FLT_MIN)));  //每个数据i的损失为-log(prob(i))
      ++count;
    }
  }
  //loss除去样本总数，得到每个样本的loss
  top[0]->mutable_cpu_data()[0] = loss / get_normalizer(normalization_, count);
  if (top.size() == 2) {
    top[1]->ShareData(prob_);
  }
}

2,反向传播

template <typename Dtype>
void SoftmaxWithLossLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
    const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom) {
  if (propagate_down[1]) {
    LOG(FATAL) << this->type()
               << " Layer cannot backpropagate to label inputs.";
  }
  if (propagate_down[0]) {
    Dtype* bottom_diff = bottom[0]->mutable_cpu_diff();
    const Dtype* prob_data = prob_.cpu_data();
    caffe_copy(prob_.count(), prob_data, bottom_diff);//把概率数据复制到bottom_diff
    const Dtype* label = bottom[1]->cpu_data();       //获得标签数据
    int dim = prob_.count() / outer_num_;
    int count = 0;
    for (int i = 0; i < outer_num_; ++i) {
      for (int j = 0; j < inner_num_; ++j) {
        const int label_value = static_cast<int>(label[i * inner_num_ + j]);//获得真实标签
        if (has_ignore_label_ && label_value == ignore_label_) {
		  //如果忽略标签，则偏导数为0
          for (int c = 0; c < bottom[0]->shape(softmax_axis_); ++c) {
            bottom_diff[i * dim + c * inner_num_ + j] = 0;
          }
        } else {
          //计算当前概率密度与理想概率密度之差（label位对应的理想概率密度为1，其他为0，故不计算）
          bottom_diff[i * dim + label_value * inner_num_ + j] -= 1;
          ++count;
        }
      }
    }
    // Scale gradient
	//缩放
    Dtype loss_weight = top[0]->cpu_diff()[0] /
                        get_normalizer(normalization_, count);
    caffe_scal(prob_.count(), loss_weight, bottom_diff);
  }
}