differential_surface_representation icon indicating copy to clipboard operation
differential_surface_representation copied to clipboard
verified publisher icon bednarikjan

Question about the overlap loss implementation

Open sagniklp opened this issue 5 years ago • 0 comments

Thanks for the great work!

The loss function in eq. 14 is supposed to encourage the area covered by each predicted patch is the same as the GT area. But according to the equation if the predicted patches have zero area the loss is zero since the second term is negative. I'm failing to understand if is this the desired behavior?

sagniklp avatar Oct 21 '20 17:10 sagniklp