YouWillLikeIt
Results
1
comments of
YouWillLikeIt
> I am the author of this paper, the KL loss shouldn't be overly minimized (this is why we set \lambda as 0.01 or 0.001). Practically, for Face Recognition task,...