It looks track2 (dped) does not use target data (tr-y) for train

[ysoo1133]

It looks Track2 (dped) does not use target data (DPEDiphone-tr-y) for train. https://github.com/Tencent/Real-SR/blob/master/codes/preprocess/create_kernel_dataset.py

There is no 'target' option in paths.yml. https://github.com/Tencent/Real-SR/blob/master/codes/preprocess/paths.yml

While, Track1 use "Corrupted-tr-y and down-sampled Corrupted-tr-y" for train as GT and input with Corrupted-tr-x. https://github.com/Tencent/Real-SR/blob/master/codes/preprocess/create_bicubic_dataset.py

Is that right? or did I miss something? If it's right then was 'cleaned tr-x' enough for target data for GAN?

Thank you for sharing the code :-) :+1:

[ttys0001]

我也有一样的疑问，因此我最后用源图像进行训练，发现效果似乎是差不多的，但是如果拿target data来生成LR，效果差很多

[cgysjy]

请问能分享一下KernelGAN的Result文件么，我总是报错，谢谢