Skip to content
New issue

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

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

Already on GitHub? # to your account

Implement SwinIR v2 #2092

Merged
merged 2 commits into from
Oct 10, 2022
Merged

Conversation

C43H66N12O12S2
Copy link
Collaborator

This PR implements SwinIR v2 from https://github.com/mv-lab/swin2sr. SwinIR v2 is SwinIR but with Swin Transformer V2 from Microsoft.

To use this, download Swin2SR_RealworldSR_X4_64_BSRGAN_PSNR.pth from the aforementioned repo, and rename it to have the .v2.pth extension.

The actual line addition of this PR is 28 lines, the rest is the model architecture.

Examples:
Original Image
00026-941083205

SwinIR
00001

SwinIR v2
00002

In this example, SwinIR v2 avoids the blocking artifacts of v1, and also respects the (artistic?) blur/bokeh of the original image.

@chekaaa
Copy link
Contributor

chekaaa commented Oct 9, 2022

Its nice that it kept the DOF , at least in this example

@Ehplodor
Copy link

That is interesting and clearly better. However I just want to point at the fact that there is still a lot of artefacts. Better, yes, definitely, but not so much IMHO.

Thank you anyway.

@Foul-Tarnished
Copy link

BSRGAN would be cool too
I like that it doesn't try to be overly sharp

@professorDeveloper
Copy link

wow its best technology

Copy link

@mphetens mphetens left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pliki scalone, poprawione

ifsheldon pushed a commit to GempollAI/stable-diffusion-webui that referenced this pull request Dec 17, 2024
Ensures `should_refresh_model_loading_params()` is called when needed. Improved code clarity.
# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants