color calibration DT4.0

Hi,
the color calibration module behaves for me a little differently in version 4:

  • the illuminant option “detect from image surface” hardly provides usable results. In version 3.8.1, the results were often similar to the white balance of the camera. Now i often get also a complete blue screen.
  • the option “detect from image edges” always produce a black screen and undefined CCT. In version 3.8.1 I often had good results with it.

And I have a small irritation with the color balance rgb module:
Is it possible that only for this module the reaction time for a double click has changed? Resetting a value with a double click is only possible with this module after several attempts.

I work with DT4.0 on a Windows 11 system.
Greetings
Thomas

Hello,

Do you have OpenCL enabled?
If yes, does disabling it help?

Kofa

Hello,
yes, OpenCL is enabled. But there is no difference when it is disabled.
It is then so slow that you can see that the option “detect from image edges” gives a good result and displays briefly before the images goes black.
Thomas

Can you upload a problematic image, and the XMP sidecar, to this forum?

Hello,
here is a sample image with the XMP sidecar.
(i hope this is ok to upload such a big file, i am new here)
I have the error with the black screen with option “detect from image edges” with every picture.
Thomas

_1BB2552.NEF (49.8 MB)
_1BB2552.NEF.xmp (11.9 KB)

Can’t reproduce on Linux (Ubuntu 20.04) with 4.1.0+26-g407f53c46. Possibly OS specific?

I’ll try with Linux and Windows when I get home.
It’s OK to upload large files. If you wish, you can support the site with donations - but this is by no means compulsory. See PIXLS.US - Support

I can reproduce this behaviour under win10. It occurs with or without OpenCL.

It looks as if it differs from image to image. Sometimes "detect from image edges” leads to a black image, sometimes "detect from image surfaces”, sometimes both. As @bildbaendiger describes in his last post it seems as if a result is produced but is somehow overwritten by a black screen.

Maybe a darktable -d all out put to see what’s going on. I’m on the cell too, so i cant test your files at the moment.

Log snippet:
darktable-log.txt (199.8 KB)

Update:
I have now installed Ubuntu 22.04 on the same computer and i am running DT4.0.0-1.1.
The error does not appear under Ubuntu. It definitely seems to be a Windows 10 / 11 problem.
I am really impressed with what a great speed darktable runs with Ubuntu. I never thought that the difference could be so big on the same machine. Amazing.
Thomas

I was able to replicate on Windows 11 darktable 4.1.0+19~g387e05538 using your image and some of mine too.

This seems to be a suspect area:
validation check failed in _iop_validate_params for type “float”, field: aspect
validation check failed in _iop_validate_params for type “dt_iop_borders_params_t”
validation check failed in _iop_validate_params for type “float”, field: g2
validation check failed in _iop_validate_params for type “dt_iop_temperature_params_t”

Do you want to start a github issue?

Thank you g-man for the infos.
I do not know, how it works with a github issue. I will first see how this works.
Thomas

I started it this morning.

image

Interesting… if you set the working profile to srgb or linear rec2020 I see issues…but at least for this image linear prophoto or linear rec709 did not induce artifacts… before this I tried several other images and did not notice this effect…

EDIT:

Hi @priort, sorry can not reproduce this under Win10. Same effect as before independent of working profile (dt 4.0).

Makes me wonder if its hardware based…I don’t see any of the reported issue as noted…I have a 12 gen intel processor Nvidia card and up to date Win 11 install. I only noted the workspace thing as the xmp of the OP had be set to rec709 and I thought that might be some of the issue…when I started playing with them I noted the difference on my system but even that was not as noted and only seem to go sideways with the other Auto WB mode… so no idea or help on my end…

I get it on Win10, 64bit Intel Dell pc. I will add specs for graphics card when I am home

On my computer (Dell 64bit Intel i7, 32GB, Intel UHD Graphics 630 and NVIDIA GeForce GTX 1060 with driver 516, Win 11 823) I have the error regardless of the input profile, it happens with every working space. Version 3.8.1 and Version 4.0 on Linux did not have this error on the same machine.

I think it is a windows only issue. The history shows a NaN (not a number) after it runs the ai analysis.

I entered an issue in GitHub, but we don’t have a lot of windows developers. I use the picker to find the white balance instead of the ai.