Image-Super-Resolution-via-Iterative-Refinement
Image-Super-Resolution-via-Iterative-Refinement copied to clipboard
Training on own dataset for (128x128) --> (512x512) fails miserably !!
trafficstars
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
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?
I have the same question, how to fix it up?
Try to lower the learning rate? I have adjusted to batch_stize4, lr2e-5
I could not fix it. So I gave up !!