Unable to reproduce the result of SRGAN_x4-SRGAN_ImageNet.pth.tar

Lornatang / SRGAN-PyTorch

A simple and complete implementation of super-resolution paper.

Apache License 2.0

410 stars 105 forks source link

Unable to reproduce the result of SRGAN_x4-SRGAN_ImageNet.pth.tar #86

Closed chenhsiu48 closed 2 months ago

chenhsiu48 commented 11 months ago

Hi,

I followed the instructions to train SRResNet first (90 epochs) and then SRGAN (18 epochs). However, I can not reproduce the result of SRGAN_x4-SRGAN_ImageNet.pth.tar. The result I get is a very blurry image. Can you give me some clues? Why do I need to download DiscriminatorForVGG_x4-SRGAN_ImageNet.pth.tar? Use it as pretrain model for SRGAN?

Fanzzzd commented 8 months ago

I trained SRResNet model without pre-trained data. Then I used this model to generate a new image then I get some results like one this: sr_comic

Lornatang commented 2 months ago

There is currently an optimizer bug, which has been fixed