search

XPixelGroup / BasicSR

Open Source Image and Video Restoration Toolbox for Super-resolution, Denoise, Deblurring, etc. Currently, it includes EDSR, RCAN, SRResNet, SRGAN, ESRGAN, EDVR, BasicVSR, SwinIR, ECBSR, etc. Also support StyleGAN2, DFDNet.
https://basicsr.readthedocs.io/en/latest/
Apache License 2.0
6.64k stars 1.17k forks source link

Model training scale: 2 ESRGAN #276

Open wolfam0108 opened 4 years ago

wolfam0108 commented 4 years ago

Hello! Can I train an ESRGAN model at x2 scale? This requires a trained model for x2. Where can you get such a model?

xinntao commented 4 years ago

@wolfam0108 Sorry that we have not provided models for X2. You can modify the network structure and train it by yourself.

wolfam0108 commented 4 years ago

Thanks for the answer. Is there a description of how to change the structure of the ESRGAN network? Or at least in what direction to dig?

xinntao commented 4 years ago
  1. A reference: https://github.com/xinntao/BasicSR/blob/master/basicsr/models/archs/srresnet_arch.py
  2. Since ESRGAN is already very large and slow for X4, it is better to first downsample the feature maps, and then upsample for 4 times, when applied on X2 scaling factor. (This is called UNet structure. )
UESTCrookieLI commented 2 years ago

hello wolfam0108 did you complete the x2ESRGAN work? I am going to run ESRGAN when scale=2, but i got some problems.