Image artifacts with multiple masks

[jsmucr]

Is there an existing issue for this?

• I checked and did not find my issue in the already reported ones

Describe the bug

I'm encountering issues when multiple masks are used.

Here I am trying to emphasize eyes of a cat using the color balance rgb module.

I need to do that for each eye separately, as one of them is in a shade.

Now – I need to use some extreme values as well as fine-tune the masks to preserve subtle details. But even though the rest of the image is supposed to be excluded from my adjusments, clearly it is not.

I've isolated the issue on a different image. I've purposedly inflated the brillance values and tuned the mask opacity and contrast to make the artifacts appear.

The mask should not affect the petals in any way, as they're clearly out of reach and not yellow-tinted.

Steps to reproduce

1. Add an instance of the color balance rgb module.
2. Max-out its brillance values.
3. Add a mask located near a bright part of the image but in a way that the mask does not overlap it.
4. Wiggle with the contrast and opacity sliders until a bright part of the image suddenly turns white.

Expected behavior

No response

Logfile | Screenshot | Screencast

No response

Commit

Dunno. I run unstable on Windows but stable 5.4.0 in WSL Ubuntu does it as well.

Where did you obtain darktable from?

darktable.org / GitHub release

darktable version

5.4.0

What OS are you using?

Windows

What is the version of your OS?

10

Describe your system

No response

Are you using OpenCL GPU in darktable?

Yes but it does not seem to be related. OpenCL does not work in WSL.

If yes, what is the GPU card and driver?

No response

Please provide additional context if applicable. You can attach files too, but might need to rename to .txt or .zip

example.zip

[wpferguson]

Max-out its brillance values.

There are warnings that doing this will cause artifacts. IIRC the guidance is to keep it less than 25%

[jsmucr]

Well, I have two questions then:

• Why is it even possible if it causes issues?
• Why does it affect masked-out parts of the image?

[pehar1]

See discussions in https://discuss.pixls.us/t/something-wrong-with-darktable-4-1-0-git272-5a1a1845-1-clipped-highlights/32540 and #12442

[jsmucr]

I've encountered this one as well quite a long time ago and thought of it as a fairly funny bug (not a feature).

This is different, though. I can't see why would a module affect masked-out portions of an image. That's not what I'd expect to be suitable to classify as an Oh well... Let's move on. issue.

[jenshannoschwalm]

1, the yellow mask tinting might not be "pronounced" enough - you might miss the very-small yellow tint. We have the hidden option darkroom/ui/develop_mask_mix=0,300000 for such cases
2. If you "somehow" increase "brightness" in UCS colorspace to extreme "supernova" values the guided feathering filter will have problems. Also we have modules having a box filter and those might also go crazy. It's now pretty hard to get into such values but still it's possible. We could decrease the UI range (as "discussed" at mentioned refs but then people start using multiple instances ...)

So for me - begging for issues but not a bug :-)

[jsmucr]

So for me - begging for issues but not a bug :-)

Oh no. :D

I myself worked around that using the exposure module. Not quite the same thing but a similar result.

The examples above are intentionally exaggerated to make the bug more obvious. In real life, the artifacts may not be as noticeable as that. So a professional could easily end up with having a poster printed in large quantities, all prints being affected by this.

I'd really appreciate if you (or anyone else who knows how this works) would take at least a brief look at the example hands on to see what's actually going on. Or this issue ends up closed and forgotten. Or maybe in a What not to do docs section which nobody would read, because this is obviously a glitch, not a feature.

I'm ready to provide as much support as I can.

[jenshannoschwalm]

Did you read the refs? The issue was "hot stuff" when working on UCS maths a while ago when color equalizer was added. Now i haven't seen any real-world issues since then.

[jsmucr]

This is not the random shiny blobs these issues refer to.

[jenshannoschwalm]

I checked your example and i can of course reproduce. As mentioned related to extreme brilliance setting, so signals before mask blending are "extreme".

Interestingly, the artefacts don't happen with some setting for "mask contrast" - just move that slider around. So there seems to be a pending bug in masking code.