有关该模块在SR任务中的使用
Closed this issue · 3 comments
splinter21 commented
1、加上一个PA模块,更适合宽度大的网络(例如EDSR,32block256channel),还是更适合深度大的网络(例如RCAN,64channel,但是共计200block)?
2、该模块,每个网络只是加一个最有效吗?在不同block后的位置加上多个,性能增益如何?
3、位置方面,加在什么位置最好?PA-EDSR是加在所有block中间。
HarukiYqM commented
Hi,
- The attention module is independent of network structures. We choose a simple backbone (EDSR) since we want to better showcase its effectiveness. It should work in both cases.
- Adding more modules will further improve performance. More information can be found in the ablation section.
- For Pre/Mid/Post, Mid gives best results. But inserting at all 3 positions yields best overall results.
jeejeelee commented
@HarukiYqM 关于“For Pre/Mid/Post, Mid gives best results. But inserting at all 3 positions yields best overall results.”这个点,您们开源的panet 是在For Pre/Mid/Post都插入了pa模块吗?谢谢您了~
HarukiYqM commented
@HarukiYqM 关于“For Pre/Mid/Post, Mid gives best results. But inserting at all 3 positions yields best overall results.”这个点,您们开源的panet 是在For Pre/Mid/Post都插入了pa模块吗?谢谢您了~
PANet 和 PAEDSR只在Mid 插入了pa模块