Some bugfixes for LateD65 whitebalance and detail mask

[jenshannoschwalm]

1. introduce the NaN aware dt_vector_max_nan() and dt_vector_clipneg_nan() vector function variants and make use of them where required.
2. For the segmentation based HLR we also require the correction to be used in LateD65 mode.
3. Make sure the OpenCL code for VNG demosaicers writes the scharr/detail mask if requested. Not done only in dual demoiser modes as they want the mask to be written from the "sharper" demosaicer like RCD.
4. EDIT: make sure DT_ADAPTATION_CAT16 is used per default

Fixes #16280
Fixes #16286
Fixes #16282

Third commit would also be for 4.6.1

[TurboGit]

Tested against 0094, the result is not identical but the diff is not visible:

Test 0094-highlight-segmentation
      Image hlrecovery.arw
 
      Expected CPU vs. current CPU report :
      ----------------------------------
      Max dE                   : 3.15145
      Avg dE                   : 0.08140
      Std dE                   : 0.25824
      ----------------------------------
      Pixels below avg + 0 std : 90.07 %
      Pixels below avg + 1 std : 90.07 %
      Pixels below avg + 3 std : 97.91 %
      Pixels below avg + 6 std : 99.95 %
      Pixels below avg + 9 std : 100.00 %
      ----------------------------------
      Pixels above tolerance   : 0.01 %
 
  FAILS: image visually changed
         see diff.png for visual difference
         (416568 pixels changed)

Diff:

Expected:

Output:

[AxelG-DE]

I just tested this PR with regards to the adaptation only and currently there is a new "bug". The temperature is fixed at 5003 K.

• Only when resetting the module, the actual value comes
• I donno whether it has been done this way, but it would be better to figure out, why adaptation didn't/doesn't obey workflow settings

PR16290_20240211-2024-02-11_14.40.28.mp4

[jenshannoschwalm]

• Only when resetting the module, the actual value comes

• I donno whether it has been done this way, but it would be better to figure out, why adaptation didn't/doesn't obey workflow settings

Please explain. I tested here with all 4 workflow options and couldn't yet reproduce ... Any preset in color calibration used? BTW i had observed such behaviour while working on the new stuff but thought i had fixed it ...

[TurboGit]

@jenshannoschwalm : The first point (Only when resetting the module, the actual value comes) is again only when resetting history from Lighttable.

[jenshannoschwalm]

Latest commit fixed the second point in segmentation where we required valid correction data.

[TurboGit]

@jenshannoschwalm : Confirmed, test 0094 pass now with the exact same value as previously.

[jenshannoschwalm]

The first point (Only when resetting the module, the actual value comes) is again only when resetting history from Lighttable.

I see. I am not sure if this was different before :-) What we have is:

1. we don't want the module to be enabled in "legacy" or "none" mode. So we don't have data available to show some meaningful temp data
2. "reset" - better reload defaults now works.

[TurboGit]

@jenshannoschwalm : I have this issue in "scene referred" workflow too on my side.

[jenshannoschwalm]

Could you check again with latest force-push ?

[jenshannoschwalm]

BTW: commit (4) issues might very well be in 4.6 too, will check after we got it sorted out here.

[TurboGit]

This latest version works for me, @AxelG-DE can you double check too on your side? TIA.

[AxelG-DE]

Please explain. I tested here with all 4 workflow options and couldn't yet reproduce ... Any preset in color calibration used? BTW i had observed such behaviour while working on the new stuff but thought i had fixed it ...

I would have hard time to explain better than my screencast, which is very comprehensive :-D

But anyhow I have tested last commit now (19:18 UTC+1) and

• it behaves again as I would expect
• I have merely tested in scene-referred (filmic)

Thanks from my side

[TurboGit]

Thanks!