JDAI-CV/CoTNet

采用3*3卷积还能构建全局注意力吗?

Opened this issue · 0 comments

k,q,v中k和v采用3*3卷积来提取局部特征,还能进行全局建模吗?