color calibration - Brainblasting...

Happy new year,
well documentated, my IQ is not sufficient for that module. I thought, I’ am experienced user )-; Spent a lot of time with the new module - for me it is a kind of mystic.
Simple workflow needed - if that fails, then that, etc. Sometimes too magenta, to greenish, too yellow and so on. Hopefully, Glenn Butcher will present a youtube Video.
Thanks in advance, Frank

1 Like

See Aureliens video https://www.youtube.com/watch?v=U4CEN0JPcoM and the detailed documentation https://darktable-org.github.io/dtdocs/module-reference/processing-modules/color-calibration/

1 Like

Thank you. I studied the documentation several times. Although I have an abo, I didn’t recognize Aurelians youtube representation.

One of the beautiful things about this module, is on majority of images, you don’t have to know anything to get great results. In settings, processing, make sure auto apply chromatic adaptation defaults is set to ‘modern’. Then reset white balance (it should be ‘camera reference’ giving 1 tint, 6502 temp), and CAT part of chromatic adaptation will automatically give great results. Linear bradford is suitable in majority of circumstances (daylight, tungsten), so you just move on without even thinking about it. There are a few if this, then that scenarios. Here is my list:
*If LED’s or coloured lights are in shot, use cat16. Other weird scenarios like mixed lighting may also call for this.
*If you want to temper sunsets, but retain blues, try non-linear bradford.
*If cct says invalid, use custom hue/chroma, or the colour dropper (which adjusts the hue/chroma). When using colour dropper, draw a square around an area which should be neutral.
*use gamut compression if lots of colours are OOG. Use clipping indicators to check this, with softproof profile = working profile (for master edit).

There is a lot more can be done, but that is my starting point. It only seems to really break if your camera has weird coefficients. Aurelien tells you how to deal with this in the video, from about 52 min mark. The other part of color calibration is the channel mixer, which a lot of people don’t understand anyway. But ‘colourfulness’ (similar to saturation) and ‘brightness’ tabs should be pretty intuitive for most users.

3 Likes

The same here.

I have read thst it should substitute the white balance, that the old white balance module is not well suited to the new scene base and linear rgb model.

But color calibration mixes too many things and provides several tools in one, some of them i don’t know even its utility.

I hope i will lear slowly.

But I think an smooth path for the not so techy would be a good thing.

May be providing a basic mode that just lets you correct temperature and a tint equivalent, although i know aureliene does not think tint is a good model, but whatever it can be substituted for.

I am confuse with its place in the development pipe too.
If it does white balance it should be placed in the place of whire balance, just after bayer interpolation and before any color correction.
But it is placed lte in the pipe and after white balance.
I know you can use the new modern mode and it will substitute the white balance, but then i have green photos, it seem it cannot read the white balance adjustments from the camera.
I have to begin touching sliders without knwing what i am exactly doing.

I have read aurelien thread, but did not understand much, it is too technical for me.

We will have to learn step by step, but a simple model would be desired, or at least instruction on how to just get similar results to white balance.

I have a sunset (unfortunately a DNG). All of my tries with color calibration ends up with magenta/red results. For interest, if any, I can upload this. Can’t find out what’s wrong, even with Tim’s tips… I will learn the new workflow.

Only other thing I can think of is sometimes I have to hit ‘reset’ on color calibration for it to work properly. This is true if I have saved it as part of a style.
Upload it as a playraw and you can see how others handle it.

This is my understanding, may be wrong:
Classic white balance simply takes sensor values and multiplies each by a different number. This has to be done before highlight reconstruction, because if you set that to ‘clip’, blown pixels will need to be turned grey, and we have to know what ‘grey’ means (we have to set all the RGB components to the same value). Aurélien says demosaic also needs a WB. This also means more work: when you update WB, you have to re-run demosaic and highlight reconstruction, too. Then, colour calibration reads the original WB coefficients, determines the illuminant’s properties (or lets you pick new ones), and transforms the image. With the modern method, the ‘rough’ (‘camera reference’) WB never changes, so there’s no need to re-run demosaic and highlight reconstruction when you change WB in colour calibration.

This process may result in ‘out of gamut’ colours. You can use ‘gamut compression’ (which intelligently compresses the colours, taking care not to squash in-gamut colours too much) and ‘clip negative RGB from gamut’ to fix this.

The R, G, B and grey tabs are simply the same as in the ‘old’ channel mixer.

Additionally, you can selectively modify the brightness and saturation based on colour. E.g. if you set ‘normalize channels’ and drag the R, G and B sliders around on the ‘brightness’ tab, the overall brightness will remain similar, but that of the individual colours can be changed. By deselecting ‘normalize’, you are given complete control.

One interesting feature is that you can use masks with the colour calibration module (something you cannot do with ‘classic’ white balance), so if parts of the image have different illumination, you can apply adjustments on the CAT tab separately to each (you will get a warning, but if you understand what you’re doing, and apply one instance of the module to each area using masks, you can make lighting more uniform).

1 Like

As I understand the color balance module, basic use is:

  • activate the “classic” white-balance module and set it to “camera reference” WB (right-most button below the sliders)
  • then switch to the color balance module, CAT tab and adjust the white balance.

For natural light images, that should be enough.
You can play around with the other controls in that tab to see if you get a better rendering, but that’s optional in most cases.

As for why it is placed where it is in the pipe-line, see the links posted by @pehar

All the other tabs are for more advanced functions, that you can ignore for white balancing purposes (i.e. telling dt what light source was illuminating your scene).

1 Like

A bit more precise: demosaicing doesn‘t need a white balance in the usual way (compensation for different lighting), but an indication how to weight r,g,b pixel values from the sensor recording for a neutral object to receive equal resulting r,g,b values in the pixelpipe when lit with D65. This is baked in when camera support is provided in rawspeed. You can override this if you measure your camera …
The whole geometry related and further colorindependent stuff can be done before doing white balance.

1 Like

Apparently some cameras require as shot as a start in WB as they have non standard D65 WB

Is it overexposed? Click the raw overexposed button and see if the magenta highlights are. Change highlight reconstruction to reconstruct in Lch and adjust the slider to get rid of the magenta

After watching the Video: the blown reds turn into magenta (one can see this by turning highlight reconstruction off) and then pushed by color calibration over the whole image. Perhaps sunrise hast to be shot with “expose to right” and massive pushing of shadows. Don’t misunderstand, Aurelian’s thougths and work are great - I would just mention that Rawtherapee produces an acceptable output of that sunrise image out of the box without banding and magenta errors.

if Rawtherapee doesn’t have magenta stuff they might use a different whitepoint, do a different highlight reconstruction or clipping - there’s no magic around.
Without access to the image it’s just speculating:

  • try set highlight reconstruction to reconstruct in Lch (maybe color, but if the reds are clipping frst, this doesn’t help)
  • if that doesnt work then use clipping - but then you lose information
  • for some cameras the raw whitepoint might be to high (i fond this with my old Canon 7d) so reducing it might help.

A lot of stuff Rawtherapee might do under the hood - in darktable it’s up to you to decide what and how much information you’re going to ditch :wink:

1 Like

Indeed, it does.

I can’t look anywhere anymore without mentally assessing the dynamic range. IMHO this is where it starts, and where is the most “bang-for-the-buck” in the quality of the result rendition. Exposure that doesn’t clip highlights provides the best starting point for this; all the curving and tone-mapping in post is just to accommodate the initial data…

1 Like

This has nothing to do with linear or scene-referred. The old white balance doesn’t match the brain + eyes adaptation to different illuminants, since it massages sensor RGB. The chromatic adaptation works in an RGB space modelling the retina cone response, so it’s closer to perceptual response.

Magenta highlights are linked to clipped highlights reconstruction, which is not the scope of the color calibration.

The scope of color calibration is to fix the remaining mismatches between recorded data and perceptual response to the scene after the raw white-balance (old module) and input color profiles have been applied. So, you can see it as an input profile tuner.

Once the mistmatches have been fixed, you can choose to export that high-fidelity image as-is, or apply artistic corrections that will not only behave better but give more reproduceable results when you copy-paste them between images.

2 Likes

Link to raw play:

Magenta highlights are linked to clipped highlights reconstruction.
Ah, i see. “Overexposing sunsets” is kind of relic from analog times, I guess.

OK, so I had misundertood.

I had understood that if you were using the new “modern” approach, color calibration would completly substitute the white balance old module.

But I understand well, white balance is still needed and prior to color calibration, at least to do camera RGB balancing (apply the correct coefficients to your camera in a standard situation).
Then you can use the color calibration to adjust colors and adapt them to different light situations or even different lights in the scene.

Well, I have to take it easy and try it in my photos, I usually do not need a precise reproduction of colors, but I like them to seem natural.

I can see now that when you use the modern adjustment, the white balance changes from “as shot” to a neutral profile.

My photo gets quite yellowish, but I can return it to a similar look changing the illuminant to “as in original shot” (the selected illuminant was Daylight).
In this case the Bradford lineal and CAT16 give me similar results.

Why “as shot” illuminant is not taken by default?
For a newbye, I think it would give him results more close to execpet, similar to the OOC jpeg.

It just dawned on me, this is the same class of logic I invoked when I made camera profiles from unwhitebalanced Colorchecker shots. All other processing equal, the renditions produced with these profiles had more-saturated color, for lack of a better description, than the renditions with the regular profiles and multiplier white balance. LittleCMS uses a Bradford chromatic transform by default.

1 Like