mlyg/unified-focal-loss

Multiclass

c-winder opened this issue · 1 comments

Hi,

In your example use-cases, you show a multiclass example, but as far as I can tell your code only works for binary classification (forgive me if I'm wrong, I use pytorch not tf). If this is the case, could you share your implementation of using the function in the multiclass case? Guessing it would be a loop over

Additionally, how do you use the delta parameter in the multiclass case? Should it have a different value based on how common each class is or would you choose a constant value for all classes?

Thanks

tayden commented

@c-winder, wondering what you came up with to make this work for multiclass inputs? Just looping?