Color Calibration, icc, real colors

Software darktable
81

My testing wasn’t extensive, but yes, with -p matrix, the Adobe ICCs showed obviously better definition in highlights than with xyzlut. I can’t remember whether I also tested lablut with the Adobe files, but I did test that option with the profiles from the SSF above and found the same effects.

With LUT ICCs, I can see hue shifts (checked with the color picker and no tone mapper) when exposure increases. It’s as though a bad desaturation were baked into the LUT, but I think it’s actually because the input values are out of domain.

Are LUT-based profiles supposed to be better? Aside from the highlights, I couldn’t see a visual difference. I only compared a couple photos. And I imagine fixing the sample range would not fix darktable not being able to do parametric masks early in the pipeline with LUT ICCs.

82

So, it goes like this: Using SSF camera data to make a matrix profile doesn’t improve the profile over making a matrix profile with a ColorChecker target shot, tested this. In order to get the advantage of the better coverage of camera response provided by the SSF data, you need to make a LUT profile.

If the -p matrix profile from the Adobe data used the ForwardMatrix, you’re just getting a target shot profile. If it used one of the DCP LUTs, should provide a similar benefit to using SSF data IF the DCP LUT wasn’t informed by a target shot, hard to know.

The other consideration is that a ICC LUT profile is a Perceptual rendering intent file, which spreads the color differently than the relative_colorimetric intent used with matrix profiles.

1 Like