search

chaiNNer-org / spandrel

Spandrel gives your project support for various PyTorch architectures meant for AI Super-Resolution, restoration, and inpainting. Based on the model support implemented in chaiNNer.
MIT License
161 stars 12 forks source link

HMA (Current #1 Upscaling Leader) #309

Open Arcitec opened 1 month ago

Arcitec commented 1 month ago

The models were finally released 3 weeks ago! Dear God, please... support this! :D

image

image

Leaderboard Ranking:

Repo:

Models:

Arcitec commented 1 month ago

Important: The HMA code is based on HAT's code. It was a direct copy of HAT's repo, with some code rewrites and better training.

You already support HAT, so adding HMA support will be a lot easier than I expected.

However, there's problems in HMA's code. I have documented all the steps I had to do to fix the bugs in HMA's / HAT's repos to make the code run locally. I submitted a PR, so it's fixed now. Just follow these steps if you want to run it locally directly from their repo:

https://github.com/korouuuuu/HMA/issues/4#issuecomment-2421013301

I think the best path forward to figure out how to implement HMA is to do a diff between HAT and HMA's repos (possibly at an older point in time of HAT's repo since I doubt HMA used the latest HAT code), and see what part of HAT's architecture that HMA actually changed. You probably have 99% of the code already.