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

一篇来自南理工的文章

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

[深度学习从入门到女装]Spatial Group-wise Enhance: Improving Semantic Feature Learning in Convolutional Network
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 pooling得到g
4、进行pooling之后的g与原group feature进行element-wise dot
5、在进行norm
6、再使用sigmoid进行**
7、再与原group feature进行element-wise dot

相关文章: