data-efficient-gans VS ESRGAN

Code for https://arxiv.org/abs/2006.10738 found: https://github.com/mit-han-lab/data-efficient-gans

I would like to make a tiny project with GANs using some high quality pictures of a single individual. I am planning to get around 500 of these and then x-flip them, however I am not sure what model I should consider for the training. I have used StyleGan2 ADA for another project which ended quite well, but I had around 14k pictures, here now the training size is much smaller and I was therefore thinking about using DiffAugment which has seemingly promising results with just 100 images.

I used a modified version of this repo, and there's also the official NVIDIA implementation, though neither have official notebooks. You can Google 'StyleGAN2 ADA Colab' and find a few starting points that way, but wait a few hours and I can clean up my notebook and post it here!

Paper: https://arxiv.org/abs/2006.10738

There are some repos which I've looked at to add deepspeed to such as DiffAugment-stylegan2-pytorch, lucidrains/stylegan2-pytorch and eps696/stylegan2 (which is in tensorflow so it would need to be translated to pytorch as deepspeed only works with pytorch right now).

Here is the paper using GANs with few data points https://arxiv.org/abs/2006.10738

Hey, for image gans, you can use smooth data aumentation https://github.com/mit-han-lab/data-efficient-gans in case you have a reasonable sized dataset.

This seems to be based on ESRGAN which is supposed to be higher quality than waifu2x.

I downloaded https://github.com/xinntao/ESRGAN , I run it using command line, no gui.

Nerd info: Uses 3 different algorithms together: https://github.com/CompVis/stable-diffusion For generating source https://github.com/TencentARC/GFPGAN For repairing faces, mouths, eyes https://github.com/xinntao/ESRGAN For upscaling Paint.Net for manual retouching, cleanup and adjustments.

Code for https://arxiv.org/abs/1809.00219 found: https://github.com/xinntao/ESRGAN

Even if it's not a simple black and white image, you can get damn impressive results when upscaling images with GAN-based algorithms, like ESRGAN, together with a suitable pretrained model. Both are free of charge. The software can be tricky to use compared to Photoshop, though.

I use ESRGAN: https://github.com/xinntao/ESRGAN

There are two Generator architectures that are in the code, however these are actually just sitting there unused, likely from work that did not pan out because if you look at their inference code it's not used at all. Instead they directly import the plain old vanilla RRDB architecture from BasicSR yet again another Xinntao repository. Seeing the theme here.