Focal Loss Alternative

This is bit of a long shot, but I was wondering if anyone knows of a loss function published this year that did better than focal loss. It was a fairly simple loss like L = (1 - p) or something of the sort.

Poly Loss, I think there were some thread where people had tried out.

Thank you! Is there a discussion of it anywhere? The more I look at it, the less I understand as to why it would work given that epsilon can be negative.

1 Like