search

TomTomTommi / HiNet

Official PyTorch implementation of "HiNet: Deep Image Hiding by Invertible Network" (ICCV 2021)
166 stars 35 forks source link

Solution to loss explosion #13

Open fkeufss opened 2 years ago

fkeufss commented 2 years ago

Thank you for sharing your code. I am trying your code and I do find the loss explosion problem. Do you know the inherent reason of it? Is there any better solution instead of restarting training with lower learning rate every time manually?

TomTomTommi commented 2 years ago

Hi, thanks for your interest. Actually, this problem occurs frequently and deserves further study, but I have not analyzed it.

Hatermelon commented 2 years ago

Thank you for sharing your code. I am trying your code and I do find the loss explosion problem. Do you know the inherent reason of it? Is there any better solution instead of restarting training with lower learning rate every time manually?

Hello, can you continue the training normally after modifying the parameters manually? I am using the manual method to modify the loss explosion problem for the first time, why after modifying the learning rate and other parameters according to the method, the model re the first round started and did not continue for 500 epochs, the learning rate did not change according to the modifications, is it something I have overlooked? Thank you.

lyq2335458686 commented 1 year ago

Hello, when I run your code, I obviously downloaded CUDA, but why can't I call the GPU when running?