POC : Introduction to a new experimental module: "3D Colorimetric Film (3DCF)"

[Christian-B]

Hello,

@kofa

the input image may have non-spectral colours;

you are absolutely right. Some “impossible” colors were falling outside the CIE spectral locus and producing artifacts (hue shifts, aberrant saturated pixels).

To solve this I implemented st_spectral_gamut, which automatically detects these colors and smoothly rolls them back toward D50 white using a soft shoulder (same formula as the user knee), while preserving luminance Y:

Principle:

• A 360-bin angular table encodes the CIE 1931 spectral locus distance from D50 white
• The purple line is interpolated in XYZ
• For each pixel:
  1. Compute angle in CIE xz plane
  2. Look up boundary distance (with angular interpolation)
  3. If exceeded > smooth asymptotic compression toward white

Here is the result using the module’s default settings; on the left is the result with spectral compression

Greetings from the Luberon,
Christian

[kofa (István Kovács)]

@Christian-B

To solve this I implemented st_spectral_gamut, which automatically detects these colors and smoothly rolls them back toward D50 white using a soft shoulder (same formula as the user knee), while preserving luminance Y:

I think when I did my gamut-compression experiments, preserving Y did not work at all, as the initial Y value was sometimes negative. I may be wrong, the code could have been buggy… I’ve found the original post - but you don’t seem to have an issue with those spots at the light sources, so well done!

module proposal: gamut compression

A new build, tagged gc-2025-10-04-16-45, from branch gamut-compress. https://tech.kovacs-telekes.org/dt-gamut-compress/Darktable-5.3.0%2B521~ga8cbb50cde-x86_64.AppImage I’ve added a new gamut compression method, that works by scaling the xy distance to get the colour inside the targeted gamut: [xy-method] [image] The scaling works by taking the distance between the white point’s xy coordinates, and those of the colour in question, checking where the line intersects with the relevant side …

[Christian-B]

@kofa

preserving Y did not work at all, as the initial Y value was sometimes negative.

I tried to describe this risk as follows:
If y_new = 1 - cie_x - cie_z becomes negative after compression, Y cannot be preserved. I simply skip the spectral compression for that pixel:

if(y_new > 0.0f)
{
    const float S_new = y_tm / y_new;
    *x_tm = cie_x * S_new;
    *z_tm = cie_z * S_new;
}

When y_new ≤ 0, the pixel keeps its original chromaticity and falls through to the regular user knee below, which has the same soft-shoulder formula. This silent fallback produces no artifacts, it just means that particular pixel bypasses the spectral compression stage.

@kofa

I’ve found the original post - but you don’t seem to have an issue with those spots at the light sources, so well done!

Thank you very much,
I remember that post; I’ll read it again tomorrow.
All the best,
Christian

[helgeschneider (Liedmann)]

Hi Christian … oh you’re very busy with this recently , a lot of changes have been made . I lift my hat you’re investing a lot of time and effort , thx for that
I enjoy it for the most part … but find a change problematic , the addition of the detail extraction in combo with being a base addition .
From my POV … it produces artifacts which I personally do not like to see in my images , can be avoided by creating a preset .
The overall module became quite complex … and is good to use .
As a normal user I can only provide … very basic info from the practical POV.
I actually prefer to work ( right now ) with the spectral tone module …

Sadly the latest build is crashing at start up … like the latest 5.7 nightly
Happening on M1 Mac .

Regards Andreas

[Christian-B]

Hello,

@helgeschneider

it produces artifacts which I personally do not like to see in my images , can be avoided by creating a preset

I agree with you, especially when using the “perceptual brilliance” slider. I made a fix in the latest version: it measures the unsmoothed luminance after applying the “ACES curve” to restore local contrast.

@helgeschneider

I actually prefer to work ( right now ) with the spectral tone module …

Interesting. I’ll compare the default values.

@helgeschneider

Sadly the latest build is crashing at start up … like the latest 5.7 nightly
Happening on M1 Mac .

Thanks for the info. I suspect there’s a problem with the GitHub build, because this week I’m no longer getting upstream updates from Darktable’s master branch.

@helgeschneider

oh you’re very busy with this recently , a lot of changes have been made . I lift my hat you’re investing a lot of time and effort , thx for that

Thanks a lot, and yes, my nights have been short lately.

Greetings from the Luberon,
Christian

[Christian-B]

@helgeschneider

Sadly the latest build is crashing at start up … like the latest 5.7 nightly
Happening on M1 Mac .

Hello,

I’ve uploaded a new version. Since I don’t know anything about macOS, I relied heavily on the AI (opencode) for help.
Please let me know if this resolves the issue, so we can share this experience with the darktable team.

Best regards,
Christian

[helgeschneider (Liedmann)]

Working now well … with no issue .
THX for the fix Christian

[helgeschneider (Liedmann)]

@Christian-B

agree with you, especially when using the “perceptual brilliance” slider. I made a fix in the latest version: it measures the unsmoothed luminance after applying the “ACES curve” to restore local contrast.

Christian … it is not the " perceptual brilliance " slider that bothers me … it is the " detail Recovery " slider in the module that is causing unwanted artifacts ( at least for me ) .
As long as this detail recovery is creating haloing effect … for me not really usable , as I would use it only locally and not globally . As far as I get it right … a tone mapper is used globally once and not multiple times on a single image .

Cheers Andreas

[Christian-B]

Hello,
You’re right, the early versions had a halo issue. I fixed it in the latest version.
The “detail recovery” slider works by extracting local contrast from the original luminance using a guided filter that smooths the image while preserving edges. This local contrast is then re-injected into the luminance after tone mapping, with an adaptive gain that compensates for the compression introduced by the tone mapper.
Before the fix, the original luminance was measured at the module’s input, which resulted in a potentially huge discrepancy and an adaptive gain that would skyrocket. Now, it’s measured after the “perceptual brilliance” slider.

I understand your hesitation. The goal is solely to recover the details lost in the highlights. I think I can reduce the default value to 15%, especially for portraits.

All the best,
Christian

[helgeschneider (Liedmann)]

Hi Christian …
at first glance the latest version looks a lot better , in terms of local contrast creating haloing along the edges !!!
So you seem …to be on the right path
Good luck with the module work

Regards Andreas

[Terry (Terry Pinfold)]

@Christian-B you have been doing some interesting work on darktable. Now DT5.6 has been released are you intending to put in a PR request for some of these modules to be included in the 5.8 release? I hope so.

[helgeschneider (Liedmann)]

Hi Christian …
nice to see , you have added small changes with the latest version of 2026/06/26.
I wiped out the old config before i installed the latest version … just to have a fresh start . Reconfigured DT to my needs .
Not sure why , but your latest build does not contain LUA ?!
I just copied it from 5.6 … so no big deal . But would be nice if you could add LUA straight into your builds .
There are now two presets in the module , but i can´t see any visiual difference between the two … maybe i`m missing something ?
Other than that … the nice feature of in image color selection and editing via mouse wheel in the color equalizer has disappeared for a while now ?
At least it does not work on my M1 machine ?!

I hope you stand the momentary heat …

Regards Andreas

[Christian-B]

Hello Andreas,

Thank you for your user feedback.

We’re in the middle of a heat wave here in France, so I’ve taken refuge in my office to enjoy the air conditioning. This has allowed me to fix most of the bugs you mentioned.

For the 3DCF module, I’ve added a slider to boost highlights and give them more contrast. I’ve also adjusted the default values for highlight desaturation. I’ll try to find the time to provide a more comprehensive update on the latest features.

@helgeschneider

Other than that … the nice feature of in image color selection and editing via mouse wheel in the color equalizer has disappeared for a while now ?

For this project, I’ve submitted a pull request: IOP Color EQ : Add interactive editing mode for color adjustments in image by Christian-Bouhon · Pull Request #21397 · darktable-org/darktable · GitHub. Can you tell me how long you’ve been experiencing these issues?
Personally, I’ve tested it on two Linux distributions and Windows 11, and the module works perfectly.

Greetings from the Luberon,
Christian

[Christian_Pfister]

Hi Christian,
I have the same issue with color equalizer on Windows 11. It worked several versions ago but in the last 2 or 3 versions it didn’t. Currently I’m running d319f2ebae.

[Christian-B]

Hello,

@Christian_Pfister , @helgeschneider
Thank you for your feedback. I ran several more tests to reproduce the bug, and I was able to reproduce it by setting OpenCL to “Very fast GPU” and “fast mode.” I must have introduced this bug while preparing the PR. I think I’ve fixed it.

Could you please test it with today’s latest version?

Thank you for your help,
Have a wonderful, sunny day,
Christian

[helgeschneider (Liedmann)]

Hi Christian … THX for the latest version .
Good and bad news …

Interactive node manupilation in color equalizer works now , cool .
It did not work for several builds …on my side , now it is good .

AI masking does not work , as no path is created … i just see the cross of the ai masking , i do click into the subject … but nothing is happening .

Lua script is working , but there is no lua folder and no lua arc file in the config folder ???

Hope you can sort out the issues

Good luck and thx again for all of your effort

Regards Andreas

[helgeschneider (Liedmann)]

@Christian-B

For the 3DCF module, I’ve added a slider to boost highlights and give them more contrast. I’ve also adjusted the default values for highlight desaturation. I’ll try to find the time to provide a more comprehensive update on the latest features.

Nice addition , i like it it does work well .

@Christian-B

Can you tell me how long you’ve been experiencing these issues?

For a longer while now … interestingly it did work longer on the MBP than on the Mac Studio .
Can´t say when it stopped working …

Regards Andreas

[Christian_Pfister]

Yes same here, interactive color equalizer works now but AI selection doesn’t. The image is analyzed but no selection is made.

[Christian-B]

@Christian_Pfister

The image is analyzed but no selection is made.

@helgeschneider

AI masking does not work ,

Hello,
I’m working on AI. Personally, I use SegNext for masks, and I made a modification to make it easier to use.
I’ll get back to you as soon as I have a solution.
Greetings from the Luberon,
Christian

[helgeschneider (Liedmann)]

Hi Christian … just for interest , why are you using SegNext ?
SAM 2 does create more precise masks , in my opinion !

Andreas