mlyg/unified-focal-loss

Does your implementation require activation in the final layers?

sarmientoj24 opened this issue · 2 comments

sarmientoj24 commented
Does your implementation require activation in the final layers?
sarmientoj24 commented

does it work with logits or no?

mlyg commented

Hi sarmientoj24,

Yes, the way the loss functions are written assume the model has a softmax activation as its final layer.

Best,
Michael

  • 👍1