Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Code release #1

Closed
Phhofm opened this issue Feb 6, 2024 · 5 comments
Closed

Code release #1

Phhofm opened this issue Feb 6, 2024 · 5 comments

Comments

@Phhofm
Copy link

Phhofm commented Feb 6, 2024

This is an interesting project, just making this issue to please let me know when you release inference or training code :)

@LehengZhang
Copy link
Collaborator

The code is released.

@Phhofm
Copy link
Author

Phhofm commented Mar 19, 2024

Thank you :)
I trained two ATD models in the meantime just to 200k
One that handles jpg compression (18 examples) https://slow.pics/s/QAFythz1
And one that can handle just a little bit of noise, lens blur, webp and jpg compression (here at 135k iters) https://slow.pics/s/I7b9lbMj
Just wanted to show real quick, will probably release them this week. Just thought id mention / you might be glad to see other people training models on your network :)

@LehengZhang
Copy link
Collaborator

We sincerely appreciate your interest in our work.
When you release the project, I would like to be informed :)

@Phhofm
Copy link
Author

Phhofm commented Mar 22, 2024

Hey, I simply wanted to show an atd model I trained that I like:
https://github.com/Phhofm/models/releases/tag/4xRealWebPhoto_v3_atd

Details and examples are in there, basically pair-trained and handles jpg and webp (re)compression together with a little bit of noise and blur handling. I really like the outputs, and considering it can still handle these degradations to some extend, it still does well on non-degraded inputs i think.

I trained it using neosr (and created these outputs of this model using the test script there, with 'norm: true' under network_g additionally)

I also trained another atd model on otf (real-esrgan pipeline) jpg compression with resize and blur kernels but non-denoising: https://github.com/Phhofm/models/releases/tag/4xNomos8k_atd_jpg

I simply wanted to show/share these results here.

(PS also, i think they are trying to add ATD support to Spandrel. Once that is in there, atd models can be used by projects using Spandrel like chaiNNer or stable diffusion web ui . Thats all the news I have)

@LehengZhang
Copy link
Collaborator

Thank you very much!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants