Changes in noise reduction for darktable 2.7/3.0

Ok. Yes, that’s in about the same as I “measured” - I counted 3 secs wit GPU and 5 secs with CPU. LOL

Are there changes to the camera profiling to determine this parameter? If yes, do you recommend that we re-profile cameras to get best results?

No, camera profiling has not changed yet. The work to change it is started, but it is quite hard to make the fitting accurate enough, and as denoising with the parameters infered by the heuristic with old profiles works quite well, I don’t know if it is worth it to continue it. I think I will rather spend some time to try to create a blind denoising module, that would not need any profile but would adapt denoising to the noise characteristics of the image nonetheless.

6 Likes

I have made a video tutorial to explain how I use the module and all its sliders:

13 Likes

Nice. Thanks @rawfiner. That should be really useful.

I have a question concerning new denoise profiled non-local means: does this method work in (linear) RGB? I think old denoise (presets/wawelet) worked in Lab, didn’t it?

Yes, the inputs are in linear RGB. The old presets worked in RGB too, but used “color” and “lightness” blending modes

1 Like

Great, thanks for the quick reply. So old denoise, was RGB, but not linear RGB?

It was linear RGB too. Denoise profiled position in the pipeline has not changed, so the inputs are exactly the same as previously :wink:

3 Likes