Why is the loss so big in the training?

[kzq666666]

Deblurring task

The configuration I changed is as follows: BATCH_SIZE: 16 TRAIN_PS: 64 VAL_PS: 64 why is the training loss so big as the epoch is 140?

[adityac8]

Hi

You can try using gradient clipping

loss.backward() 
torch.nn.utils.clip_grad_norm_(model_restoration.parameters(), 0.01)
optimizer.step()

Thanks

[FrankLinxzx]

Hi

You can try using gradient clipping

loss.backward() 
torch.nn.utils.clip_grad_norm_(model_restoration.parameters(), 0.01)
optimizer.step()

Thanks

i meet this error too where to add this code? thx

[adityac8]

Hi @FrankLinxzx

You can add these lines here. https://github.com/swz30/MPRNet/blob/435f483b6433a6cf8d43f4df9c20eebba47587bb/Deblurring/train.py#L124

Thanks

[FrankLinxzx]

Hi @FrankLinxzx

You can add these lines here. https://github.com/swz30/MPRNet/blob/435f483b6433a6cf8d43f4df9c20eebba47587bb/Deblurring/train.py#L124

Thanks

so much thx

[bathcenter]

去模糊任务

我改的配置如下： BATCH_SIZE: 16 TRAIN_PS: 64 VAL_PS: 64 为什么epoch是140，训练损失这么大？ Hello. TRAIN PS: 64 VAL PS: 64, I know

TRAIN PS is the input picture size, Val PS does PS mean,

[adityac8]

Hi @bathcenter

Could you add gradient clipping and see if it is resolved.

Thanks