Mask used by forward and output function

[Zhenyi-Liu]

Is mask used in forward function a different one than the soft mask used in output function? (seems different)

And is there any insight on the the threshold value?

Thanks!

[wyf0912]

I did not use the KL_Loss since it seems that the results are not as good as directly using L1/MSE loss (even if I try to use a soft mask). ╮(╯▽╰)╭