Sha-Lab / FEAT

The code repository for "Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions"
MIT License
418 stars 84 forks source link

label problem for meta-training stage #46

Closed freeman-1995 closed 3 years ago

freeman-1995 commented 3 years ago

Hi thanks for the great work!

I have questions about the labels used in meta-training stage, I find you didn't use gt_label in your code. you put gt_label in cuda(data, gtlabel = [.cuda() for _ in batch]), when computing cross_entropy, you use other labels(loss = F.cross_entropy(logits, label)), which is prepared by prepare_label func, this is very weird