About Skeleton-Detail Transformerarchitecture

[Terry]

Hello, thanks for opening such an excellent job. I have a question after reading your paper. In Fig. 3. 'Skeleton-Detail Transformer architecture. It consists of a self-attention layer, a cross attention layer, and an optional globally self-attention layer'.
When I read your code, I found that the code does not seem to use self-attention layer before cross attention layer. Do I understand mistakes?

[Wenxiao Zhang]

Hi, @yugang-guo. I check the codes and find you are correct and I am sorry for this. It seems I upload the codes used for ablation study, which I comment the self-attention layers in

SDT/models/model.py

Line 519 in 29933e1

 # self.sa_coarse = SA_Layer(out_channel_size_1) 

SDT/models/model.py

Line 563 in 29933e1

 # x1_coarse = self.sa_coarse(x1_coarse) 

It should work without changes by just uncommenting these lines, but the corresponding models seems been deleted since I have graduated.

[Terry]

OK,thanks for your reply.