Darktable: Demosaic issues?

Hi all,

I am seeing some funky colours in this image. Near the bottom of the brightly lit area, I see a lot of purple dots, which i figured was a demosaicing issue. Could someone help with processing this in Darktable? What should I be doing exactly?
_DSC0802.NEF (12.7 MB)

Sounds like this: Anyone see this: Black rectangles along darkroom image view edges it’s not this
But maybe not. A screenshot would be super helpful!

And, welcome to the forum!

Thanks!

Here are some screenshots:

PPG demosaic:

Amaze demosaic:

Vng4:

But I was hoping for a much cleaner result here without all the coloured dots…

What version of dt did you use? I’m using 3.5.0+2426~g4619fa474. I downloaded your image and can reproduce the “noise”, independent of the demosaic mode I use. But I see a very different result if I use one of my own images, also taken at ISO 100 with an old Sony Alpha 77. As you can see in the screenshot (darkroom 100 %): no colored dots, even not in the darkest parts of the image.

I tried all demosaic modes available, all with nearly identical result. So for me your issue does not look like a demosaicer problem. Do you see the dots on all your images?

Basically all demosaicers have to approximate the colors for the outermost pixels (the number of lines/rows affected slightly differs between the algos, ~3 mostly) in dt 3.6 some subtle issues have been fixed.

This has nothing to do with the “Anyone see this:…” issue

1 Like

The area where the purple dots appear is covered with fibres. So you’ll have reflection causing very tiny highlights (<= 1 pixel), and possibly diffraction phenomena. Also, some of the pixels in that area are clipped.
So you end up with neighbouring photosites seeing/reporting very different intensities, and in a random pattern. That is exactly what you see in a noisy image. But here, the noise is in the subject, not the camera.

(Just for completeness: switching the camera to B&W has no influence on the raw file, afaik).

I agree with your analysis. The chromatic aberrations module fixes this nicely (I didn’t take the time to fine-tune the sliders, but as there are no other colours to protect, this is how I’d actually leave them, and then mask the module if I wanted to limit the effect):

3 Likes

What you see is exactly why demosaicing is a “tricky buisiness” :slight_smile: For noisy input some demosaicers do pretty good (in dt there are AMaZE and RCD) results, some are not suited as good for such input, PPG and VNG both tend to overshoot and thus produce the color spikes.

dt has the rawdenoise module that can help for this problem as it works on the raw data and possibly deliver less noisy data for demosaicer - though reducing sharpness and desat as a tradeoff. So - only to be used in special cases.

3 Likes

Here’s how the picture looks in NX Studio:

I don’t know anything about that NX software :slight_smile:

darktable is not a solution to do a quick-shot. It also doesn’t try to develop your image with one click. Instead it’s a tool to give you a change to get your image “perfect” but will ask you to take a learning curve. So to get you into it

  1. you can see that some sensor data is overexposed, (find the button on the lower right), as that is in a region with lots of detail (the texture of the floor) the color estimation for the demosaicer is mislead.
  2. the whitebalance of this image also amplifies the rgb channels quite different adding to the problem in 1)

I don’t know how NX works at all, maybe it knows about some special things when the camera needs further help. In dt you would first a) maybe use highlight reconstruction in lab mode? b) do color smoothing in the demosaicer c) …

2 Likes

Highlight Reconstruction in LCh made all the difference!

Thanks for all the feedback folks!

Nice to hear. Just a but: dt is not for “quick hacking”! This might be a good solution for this image but it somewhat decreases quality in other ways. So this is not the default option in dt. Keep going!