Introduce the Result Into Self-Attention

Chengcheng Ye

Traditional self-attention mechanisms in convolutional networks tend to use only the output of the previous layer as input to the attention network, such as SENet, CBAM, etc. In th…