search

Alibaba-MIIL / ASL

Official Pytorch Implementation of: "Asymmetric Loss For Multi-Label Classification"(ICCV, 2021) paper
MIT License
732 stars 102 forks source link

A question about the shifted probability? #45

Closed ccc-pno closed 3 years ago

ccc-pno commented 3 years ago

Sorry, may I ask you a question？

In paper, you define the shifted probability, pm, as:
pm = max(p-m, 0) But in code, you define it as: xs_neg = (xs_neg + self.clip).clamp(max=1) I think it is means: pm = min(1-p+m, 1)

Why ?

mrT23 commented 3 years ago

https://github.com/Alibaba-MIIL/ASL/issues/10

ccc-pno commented 3 years ago

10

Thanks！