Tendency map of batch loss

[sxd0071]

Hey, thanks for your sharing. It seems that there isn't tendency map of loss or psnr of evaluation to monitor the training process, and how can I know when is proper to stop the training. Look forward to your reply

[Paper99]

Instead of applying an early stopping strategy, choosing the biggest evaluated PSNR value during 1000 training epochs is our preferred way. The reason is that the PSNR value still has a small increase in the later (800-1000) epochs.

[sxd0071]

Hey, thanks.That is to say, we can ignore the psnr after 1000 epochs, which just has some small increases and is not neccessarily be cared about. Is this understanding right?

[Paper99]

Yes, you're right.

[sxd0071]

Thanks.