I tried to adjust white balance via Color Calibration module of Darktable 3.6.0.1. and the new feature by using a chart. I have Datacolor Spyder Checkr chart with 24 colors. The profile quality report shows a “good” result, independent of the CAT method, i. e. as shown by the attached files (screen shots), even if the whole picture looks greenish. The other screen shot shows the result, in case I used color picker for CAT adjustment. The calibration result with the color checker seem to be independent of the CAT processing. My screen is calibrated using Datacolor Spyder.
I guess I make a mistake in thinking and reasoning. My goal is to have the best match of colors and while balance for my pictures on screen and later as prints. I appreciate any help and advice how to achieve this.
Looks wrong to me…be sure to compute in none neutral and lowest average and take the one with the smallest delta E…then be sure to use the check mark to apply the setting…also I think the values are valid after adjusting for the exposure and BP compensation listed after the calculation…perhaps share the raw file for others to test if you are still having issue…
I attached the RAW file. What I cannot understand is that the color calibration via chart is not effected by the CAT settings. I can deliberately mistune the colors by CAT settings. Still remains the profile quality good.
That observation is prior to accepting the computed profile. Once I store those settings of color calibration module as preset I can apply them without problem. _DSF5861.RAF (23.2 MB)
At first, the optimal parameters are computed but not applied. You get a quality report about the fitting that lets you decide if you are going to accept or not these settings. At this stage, the profile is not yet a setting but a virtual proposal. Take that as a quote from the dt engineers that let you the opportunity to refuse the service.
Then, you hit the “accept settings” button if you are happy with the quality, and the virtual settings are actually applied for real.
Imagine you already had ok-ish settings, if the computed profile is bad and applied straight after being applied, you loose everything. So, it’s just a double-opt-in process.
This is what I get…looks pretty good to me…if you start to edit the settings above in the cat tab then you are no longer using the colorchecker from what I see…I just ran default ColorCalib and then ran the colorchecker and then used the recommended EV and BP comp correction to correct in the exposure module and reran the optimization and then applied it…
So to use this you need a style with the same color profiles and the exposure settings for this to be accurate…
Your target shot has some shadowing in the lower right compared to the upper left so to be even more accurate you might want to try and watch for that…
thank you for your explanation. I understand your reasoning that it is a virtual proposal. I still have the difficulty in understanding the process, say workflow.
Assume the profile quality report based on a color checker states it is bad. How can I improve this? CAT settings do not influence that analysis. I thought I can fine tune the colors using CAT and then recompute the profile to get a better result.
thank you for showing your processing of the image. Indeed you can adjust the settings in CAT tab, so that you may not need the calibration with a color checker. But then, what is the added value by this calibration help? In fact, you cannot intervene to improve the quality of the calibration by a chart.
Another question: You wrote you used the recommended EV and BP compensation correction of the exposure module. If I select automatic mode in exposure module, the whole image becomes really dark as I can watch this from histogram, too. In exposure module I use scene-referred default as preset.
As mentioned in the chapter about color calibration : The settings used in color calibration depend on the chosen CAT space and on any color settings defined earlier in the pipe within the white balance and input color profile modules.
This means that you need to select and set an adaption before you can start to calibrate. If you calibrate and change adaptions the colours will shift/be “off” (see end of the attached video).
These are the EV/BP settings needed to determine the correct colour calibration, not necessarily for the images you are going to apply it to later on. I use the colour picker on the white patch in combination with the exposure module to reach ~95 (L channel of L*a*b] to get a reasonable starting point (not necessarily needed, but I like to do it this way).
Be aware that the Normalization values can get you into a sort of loop (+1, -1, +1 …) so getting really close is “good enough”. Spot on isn’t needed, do try a few iterations though.
Here’s a short and simple video showing the process:
Now you have a colour correct base and can create a preset/style from this color calibration setting and apply it to the images shot in this particular lighting/colour setup. The adaption is also applied (it is part of the module), but the rest (exposure for example) isn’t so you can use your (default) scene-referred settings and editing practises.
With this base you can be artistic and do colour grading/manipulations if you want/need to.
Interesting …the only way I get the illuminant that you do is if I have the WB module set to as shot not reference and then I get the double WB error. Not sure how you get this. I even went to legacy WB and no processing in the preferences… I get similar delta e values but way different illuminant??
This is what you get…
When I try using non reference WB (as shot) I get similar to you but an error…
I’m assuming this is because I’m not playing by the out-of-the-box rules
My auto apply pixel workflow is None and my auto apply chromatic adaption is Legacy
If I use darktable I do use the scene-referred workflow but I do not like the defaults at all and set them manually. Most of the time I do switch to the new chromatic adaption too, but I check both, starting with the legacy one.
But the above is rather simple and straightforward (as can be seen in the history) when I start:
No filmic, no exposure, legacy WB adaption. Only things auto applied are: Lens correction (I do always recheck this), highlight reconstruction turned off and demosaic set to either markensteijn 3-pass+VNG (+greens full average) or RCD+VNG (+greens full average). This is my base line.
The only 2 modules that are added during this calibration procedure are exposure and colour calibration. And I do have to set the adaption manually (choose CAT 16).
All the above seems to be irrelevant: I just switched to the defaults (scene-referred + modern) and, redoing what I did in the video, I get this:
As you can see I had to go back/forth to get the exposure correct (as suggested by the Normalize values). Only thing I did not have to do was set the adaption, which is part of the auto applied scene-referred thingy.
There is a small difference in certain colours between them. Which I think is explainable: In both cases CAT16 is used, but the automated version does a bit more tuning and is probably just that bit more precise.
Hi Jacques,
your video is very useful. Finally, I do understand how the calibration via chart functions. I assume the white balance module was set in your case to camera reference as default. In the color calibration module you use CAT16 adaptation with “same as pipeline (D50)”. At this step the white balance of the picture does not look good. But, after the calibration via chart (exposure compensation and black offset as an intermediate step) and accepting the result suddenly corrects the white balance.
I tried the same procedure to the image and I got very similar result like your second try, while in Darktable preferences, under processing “auto-apply pixel workflow defaults=scene-referred” was set in my case.
Besides, I had the difficulty to set the brightness for the background of a series of pictures. Using the color picker and Lab space it is practical to set L value to a certain value by changing the exposure.
Thank you very much for taking your time and illustrating how the color calibration works. I have to get more experience using Darktable, I notice. It is a very powerful tool.
How do you not trigger the double WB warning that was my question…I assumed you were using none as a workflow so I did the same but never got the blue illuminant that you had unless I had the double WB error…
This was the part I could not get in your earlier example. If you simply do this you should trigger the double WB warning unless you change legacy WB to Camera reference??
Your second attempt looks more like what I got by any combination or means that did not trigger the WB error…
Provided color calibration and white balance modules both are active and you select in white balance module other than “camera reference” then the “WB applied twice” message in red appears.
There’s an option in the (processing) preferences: show warning messages. Guess what it is set to for me
This from the manual:
CAT warnings
the color calibration module will show warnings in the following circumstances.
If the color calibration module is set up to perform chromatic adaptation but the white balance module is not set to “camera reference”, warnings will be shown in both modules. These errors can be resolved either by setting the white balance module to “camera reference” or by disabling chromatic adaptation in the color calibration module. Note that some sensors may require minor corrections within the white balance module in which case these warnings can be ignored.
The italic highlight is mine.
I’m assuming (wrongly?) that, because you need the color calibration module to do the colour checker process you can ignore this warning in this case.
Not sure if it is still the case but you use to get these warnings all the fricking time when using legacy and using the color calibration module (not to do WB, but to do “channel mixer” stuff). That’s the reason the preference option was introduced if I’m not mistaken.
Thanks I did not know about that…I need to review all the preference settings again…as far as I can see you get the exact same WB result if you use legacy as shot or modern and in the Cat drop down choose as shot
