Training on own dataset for (128x128) --> (512x512) fails miserably !!

[ManuBN786]

I was training on my own dataset for (128x128) --> (512x512)
I have attached the train logs :
train.log
val.log

The tensorboard visualizations look like this:

And the losses & psnr aren't improving

Please help me to fix it

[guangying121]

Hey, I met with the same problem, but, but my l_pix is normal like 0.05, my sr_img and PSNR are the same situation like yours. Did you solve it?

[mgl666]

I have the same question, how to fix it up?

[tA312]

Try to lower the learning rate? I have adjusted to batch_stize4, lr2e-5

[ManuBN786]

I could not fix it. So I gave up !!