前向传播
template <typename Dtype>
void SoftmaxLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
const vector<Blob<Dtype>*>& top) {
const Dtype* bottom_data = bottom[0]->cpu_data();
Dtype* top_data = top[0]->mutable_cpu_data();
Dtype* scale_data = scale_.mutable_cpu_data();
int channels = bottom[0]->shape(softmax_axis_);
int dim = bottom[0]->count() / outer_num_; //dim表示要分类的类别数,count()得到的是总共的输入Blob数,outer_num_得到的是是每一类的Blob数
caffe_copy(bottom[0]->count(), bottom_data, top_data); //先将输入拷贝到输出缓冲区
// We need to subtract the max to avoid numerical issues, compute the exp,
// and then normalize,减去最大值,避免数值问题,计算指数,归一化
for (int i = 0; i < outer_num_; ++i) {
// 初始化scale_的data域为第一个平面,其中scale用来存放临时计算结果
caffe_copy(inner_num_, bottom_data + i * dim, scale_data);
for (int j = 0; j < channels; j++) {
for (int k = 0; k < inner_num_; k++) {
scale_data[k] = std::max(scale_data[k],
bottom_data[i * dim + j * inner_num_ + k]);
}
}
// 输出缓冲区减去最大值
//a_k = a_k - max(a_i)
// C = alpha . A . B + beta * C
// A为单位1矩阵
caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, channels, inner_num_,
1, -1., sum_multiplier_.cpu_data(), scale_data, 1., top_data);
// exponentiation
caffe_exp<Dtype>(dim, top_data, top_data);
// sum after exp
caffe_cpu_gemv<Dtype>(CblasTrans, channels, inner_num_, 1.,
top_data, sum_multiplier_.cpu_data(), 0., scale_data);
// division
for (int j = 0; j < channels; j++) {
caffe_div(inner_num_, top_data, scale_data, top_data);
top_data += inner_num_;
}
}
}
后向传播
template <typename Dtype>
void SoftmaxLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
const vector<bool>& propagate_down,
const vector<Blob<Dtype>*>& bottom) {
//获得data、diff指针
const Dtype* top_diff = top[0]->cpu_diff();
const Dtype* top_data = top[0]->cpu_data();
Dtype* bottom_diff = bottom[0]->mutable_cpu_diff();
Dtype* scale_data = scale_.mutable_cpu_data();
int channels = top[0]->shape(softmax_axis_);
int dim = top[0]->count() / outer_num_;
caffe_copy(top[0]->count(), top_diff, bottom_diff); //先用top_diff初始化bottom_diff
for (int i = 0; i < outer_num_; ++i) {
// 计算top_diff和top_data的点积,然后从bottom_diff中减去该值
for (int k = 0; k < inner_num_; ++k) {
scale_data[k] = caffe_cpu_strided_dot<Dtype>(channels,
bottom_diff + i * dim + k, inner_num_,
top_data + i * dim + k, inner_num_);
}
// 减值
caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, channels, inner_num_, 1,
-1., sum_multiplier_.cpu_data(), scale_data, 1., bottom_diff + i * dim);
}
// 逐点相乘
caffe_mul(top[0]->count(), bottom_diff, top_data, bottom_diff);
}