gan and other losses are nan

Question

gan and other losses are nan

tangeroo opened this issue 5 years ago · 2 comments

@avalonstrel Hi,
Thanks for your excellent work!
When I train this net with my own data, I tried many times,also made learning rate smaller,Nan of losses happened soon or later.By the way,My mask configuration is irrmask/random_free_form.Do you have any suggestions for this promblem?
Thanks.

Answer 1 · 2020-08-04T13:15:37.000Z

Hi, I come across the same problem. Can you tell me how to solve it?

Answer 2 · 2020-08-05T01:31:55.000Z

@WangQinghuCS
hi, in my case , it's because my dataset has some problem. My net input is calculated by (img-min)/(max-min)-1, however some images their max = min, so the input is nan. Maybe you can check your dataset.