zhenghuizhao / transwcd Goto Github PK

Dear author, I have carefully read your public code, but the training script does not seem to involve the Dilated Prior Decoder and Labeled Gated Constraint mentioned in the paper. In addition, the code uses the pre-trained model parameters of SegFormer and relies on additional information, which seems to be contrary to the paradigm and principles of weakly supervised learning. I hope the author can answer these two questions in his busy schedule.

Thank you very much for your excellent work!!!

One problem was found during code execution。

cam_to_label function only has four parameters, but in train_transwcd.py convert five parameters to this function.