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

Low light enhancement #288

Open
Euanggelos opened this issue Jul 9, 2024 · 3 comments
Open

Low light enhancement #288

Euanggelos opened this issue Jul 9, 2024 · 3 comments

Comments

@Euanggelos
Copy link

Hi
Could you include support for some of the models included in this link?
https://github.com/caiyuanhao1998/MST-plus-plus
specially for higher PSNR? Restormer MPRNet MST-L MST++

@joeyballentine joeyballentine transferred this issue from chaiNNer-org/chaiNNer Jul 9, 2024
@RunDevelopment
Copy link
Member

We should already support Restormer and MPRNet.

About MST and MST++: These are meant to reconstruct for full spectrum from an RGB image, AFAIK. So given an RGB image, they will output an approximation of the light frequency spectrum for each pixel. I guess this could be used for Low-light enhancement, but not directly as is.

So I don't know which specific models you want us to support. @Euanggelos, could you please link them more directly?

@Disonantemus
Copy link

We should already support Restormer and MPRNet.

I tried both today from spandrel Readme links, using ChaiNNer:

  • Restormer: No Low-light model; I did test 4 (3 worked, from 13), but did other things like: deblurring, denoising. I did find restormer.pth, did load, but finish with an error.
  • MPRNet: No Low-light model, I did test 2 models (from 3) and worked, as expected doing deblurring and denoise. I did find mprnet.pth, but doesn't load as a model.

Maybe both doesn't have Low-Light Enhancement models, and are in the Results on NTIRE 2022 HSI Dataset just to compare older models.


From NTIRE 2024 Challenge on Low Light Enhancement, seems that RetinexFormer and MST++ are nice aditions, because they ranked top-2.

From Readme, RetinexFormer is supported, but models from links says unsupported in ChaiNNer. Maybe ChaiNNer (0.24.1) doesn't have an updated spandrel that support this.


In summary, I didn't find any Low-Light models in ChaiNNer that works.


chainner-bin (Arch AUR): 0.24.1-1 (Last Release)
distro: Arch Linux x86_64
kernel: 6.6.22-1-lts
shell: bash 5.2.26
term: tmux
cpu: Intel i7-4790 (8) @ 3.600GHz
gpu: AMD ATI Radeon RX 470/480/570/570X/580/580X/590

@Disonantemus
Copy link

  • Maybe we need to wait for next update (Spandrel and/or ChaiNNer)?
  • Is possible to compile Spandrel to get/test newer commits? (I don't know how, not a programmer)

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

3 participants