Closed wjc0602 closed 2 years ago
你好,我想问一下你在论文中提到的特征分析中,是用什么方法将注意力绘制到原图上的?就是在Attention操作之后,得到的特征图是多通道的,你是用什么方法将其合并的?
你好,attention map最后是只有一个通道的,实现方式见:https://github.com/pengzhiliang/Conformer/issues/10#issuecomment-903207037
好的 谢谢
你好,我想问一下你在论文中提到的特征分析中,是用什么方法将注意力绘制到原图上的?就是在Attention操作之后,得到的特征图是多通道的,你是用什么方法将其合并的?