Spatial Group-wise Enhance: Improving Semantic Feature Learning in Convolutional Network

文章的思路很简单,类似于SENet(对channel做attention)、spacial attention就是将channel分为group, 然后对每个group进行spatial的attention 模块如上图所示, 1、将feature map按channel维度分为G个group 2、对每个group单独进行attention 3、对group进行global average pool
相关文章
相关标签/搜索