Issues with PSA refinement and final segmentation stages

[yijiangWang0316]

Hello author, I read your paper in detail and thank you for your work.
But I found two things in the code that cannot correspond to the description in the paper:

1. PSA refinement. You use ACR* for training. Where is the code? How to use transformer to train PSA?
2. In the final semantic segmentation stage, should ResNet or ViT be used as the backbone training segmentation network?

Looking forward to your reply！