It seems to have no effect on this work.

[MoonWolf9067]

@joe-siyuan-qiao Hi,thanks for your work! When I use the following code in my work, it seems to have no effect . Before the convolution, I made changes to the convolution kernel according to the following code.Then the convolution result is sent to the group normalization. So, can you give me some ideas? Thanks

[csrhddlam]

Hi, thanks for your interest! It is a little tricky to implement WS in Tensorflow. You need to make sure that WS is executed correctly in both the forward and the backward pass. If you're using the latest version, we recommend adding those lines before this line. We also have an example of using WS in dgcnn. Moreover, you could use tf.custom_gradient and/or tf.print to further verify your implementation. Thanks.

Huiyu