xuguodong03/SSKD

Implementation of correct number of samples

Closed this issue · 1 comments

Hi, I look your implementation of counting correct number of augmented samples in student.py

254-tmp = torch.nonzero(rank, as_tuple=True)[0]
255-correct_num = tmp.numel()

Why I think correct_num represents the incorrect samples?
Thanks!

Thank you for pointing the bugs!
tmp.numel() does represent the incorrect samples. This may be the typo when I re-organised the code. And I have fixed it.