Allow custom LRs to train on JPEG artifacts, dithering, etc

andreas128 / SRFlow

Official SRFlow training code: Super-Resolution using Normalizing Flow in PyTorch

Other

832 stars 111 forks source link

Allow custom LRs to train on JPEG artifacts, dithering, etc #11

Open n00mkrad opened 3 years ago

n00mkrad commented 3 years ago

Currently, ESRGAN outperforms SRFlow when it comes to uncompressed upscaling.

And SRFlow can't be trained on JPEGs or other artifacts.

Is it possible to use custom LR images that have been augmented?

ALLinLLM commented 3 years ago

And SRFlow can't be trained on JPEGs or other artifacts.

Why？

n00mkrad commented 3 years ago

Why？

Because the dataset preparation script seems to directly use downscaled versions of the HRs, not allowing any manual editing.

martin-danelljan commented 3 years ago

You can train SRFlow on any kind of LR images, eg with JPEG, noise, etc. Please just modify the data processing pipeline to include the degradations you want.

Regards, Martin