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

Results coming from versions 1.X were sharper #200

Open
FrankEscobar opened this issue Feb 27, 2024 · 5 comments
Open

Results coming from versions 1.X were sharper #200

FrankEscobar opened this issue Feb 27, 2024 · 5 comments

Comments

@FrankEscobar
Copy link

FrankEscobar commented Feb 27, 2024

Hello all and thank you for your work Attila!

I've found that images denoised with older versions were sharper, is that something that other people has notice that too?

1X
2X

Comparative:
comparative

@atafra
Copy link
Collaborator

atafra commented Feb 27, 2024

Hi. No, in general this shouldn't be the case but exceptions like this can always happen. In general, quality has been improved since 1.x, which can be measured using various quality metrics as well.

The filters in OIDN have been updated multiple times in both 1.x and 2.x versions, so it's unclear which actual versions you tested. Also, are you pre-filtering the albedo and normal buffers to maximize sharpness and image quality?

In this particular case, it's also a bit hard to say which version is actually closer to the ground truth. The additional sharpness in your 1.x image looks a bit like residual noise.

@FrankEscobar
Copy link
Author

Since my buffers were already pretty clean it makes no difference

@atafra
Copy link
Collaborator

atafra commented Feb 27, 2024

If this is the case, are you using the cleanAux filter option?

@FrankEscobar
Copy link
Author

This was with cleanAux activated that makes a bit better result that without cleanAux.

The weird thing is that this is coming from a wall detail but the floor has more detail than before.

@atafra
Copy link
Collaborator

atafra commented Feb 28, 2024

It's actually not that weird because even though the overall quality is higher, it's not possible to guarantee that in all cases there won't be any slight variations in quality. This seems to be such a case.

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