It does so by squeezing a C*H*W feature map into a C-dimensional vector via pooling, followed by a linear layer and activation function that reduces the size to ...
確定! 回上一頁