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

train ESRGAN and loss output ：l_d_real: 3.2484e-01 l_d_fake: 5.3701e-01 D_real: -1.0425e+01 D_fake: -1.2688e+01 #160

Open qingfengmingyue opened 5 years ago

qingfengmingyue commented 5 years ago

train ESRGAN and loss output ：l_d_real: 3.2484e-01 l_d_fake: 5.3701e-01 D_real: -1.0425e+01 D_fake: -1.2688e+01 ,is ok?

yanmenglu commented 5 years ago

@qingfengmingyue I have a question to ask you, why I often see the ESRGAN model is 4x，but 2x in not often. why?

xinntao commented 5 years ago

@qingfengmingyue, It seems OK. I will upload a log example tomorrow. @yanmenglu, X4 is often used from SRGAN and many works just follow SRGAN settings. And the difference of PSNR-oriented results and GAN-based results for X4 are larger than that for X2. That is, the difference for X4 is not obvious and X4 is enough.

AloneGu commented 4 years ago

@qingfengmingyue, It seems OK. I will upload a log example tomorrow. @yanmenglu, X4 is often used from SRGAN and many works just follow SRGAN settings. And the difference of PSNR-oriented results and GAN-based results for X4 are larger than that for X2. That is, the difference for X4 is not obvious and X4 is enough.

so where's the log example now? @_@