jinhaolee / wca Goto Github PK

Thank you for your outstanding work. After running your released code (with default parameter settings), I found significant discrepancies between my results and those reported in the paper. I tried several different random seeds, but the discrepancies persisted. For instance, for the ViT/B-32 model on the ImageNet dataset, the accuracies for different methods are as follows: {'clip': 60.83, 'clip-e': 62.20, 'clip-d': 61.88, 'waffe': 61.95, 'cupl': 63.25, 'ours': 61.21}. I'm unsure whether the issue lies with the code or my execution.

Your research work is impressive!
BUT
Is there anything wrong about the link of the ImageNet-R?
Beacuse it may be same as the link of oxford-iiit-pet