maryam089 / sdvit Goto Github PK

Hi! Thank you for awesome works :)

I could reproduce the result of paper using baseline_sweep.sh but I want to enable T3A to get maximum score.

Could you explain how to enable this?

Thank you!

Hi, sorry for several requests.

Your release ImageNet pretrained T2T checkpoint (81.5_T2T_ViT_14.pth.tar) is corrupted.
I also tried to find same uncorrupted checkpoint but failed to load.

Could you share working T2T ImageNet pretrained checkpoint?

Thank you!

Thanks for your great work and sharing. I have a question to ask you. In this article, K-L divergence is used to calculate the difference between the final and randomly sampled class tokens. What I want to know is whether Domain Label is used in this process? That is to say, is it clear which domain each sample belongs to, for example, the sample belongs to Cartoon or Sketch. Looking forward to your reply. Thanks.

Hi. Thanks for the nice work!

I have a question about visualizing the attention maps. Do you use "visualize_attention.py" script? In that script, it seems that each attention head is visualized seperately. Do we need to average them? Is any thresholding used? I would be grateful if you provide the script used for producing the attention maps in the paper. Thanks