search

nabsabraham / focal-tversky-unet

This repo contains the code for our paper "A novel focal Tversky loss function and improved Attention U-Net for lesion segmentation" accepted at IEEE ISBI 2019.
357 stars 72 forks source link

loss: Nan. #29

Open linyingyingkarina opened 2 years ago

linyingyingkarina commented 2 years ago

Hi, thanks for your work.

I have succeeded to run the code. But when I ran it in a new GPU, the data, the code, all the same. However, during the training, even in the first epoch. The loss is shown with Nan. Do you know why? or any possible of causing this??

Thanks a lot.

noamsuissa commented 1 year ago

Having the same issue. Any success lately?

xiongjhang commented 1 year ago

Same issue. This can be avoided when gamma is set to 1.