Originally Posted by N3W813
Yes, my black has also been raised from 0.0134 to 0.0145 fL with 3DLUT enabled. Like you said, you can lower brightness until black is lowered to the initial value but you will raise the gamma at the low end a bit. I think you will just have to pick your poison for now until Graeme can research into why black is raised slightly.
I've looked into this sort of thing a number or times, and as far as I can determine, it seems to a result of profile inaccuracy. Plasma displays in particular seem to have an extremely low black point, and both the instrument and profile accuracy are a challenge.
[Technically what happens is that RGB 0,0,0 is looked up in the forward table, and then that CIE/Jab value is reverse looked up, and arrives at a non-zero RGB value. This is expected when the display has a "dead band" at zero. If the profile is not sufficiently accurate, then there may be a slight error from the true device black value. By normal display standards we are talking about a very, very small error - the above it is 0.004 cd/m^2, way below the expected accuracy of the instrument. Compared to a full scale of (say) 100 cd/m^2, that's an inaccuracy of 0.004%! It's miraculous that it is so small. I'm not clear under what circumstances it will be visible, other than observing the display in pitch dark with a black raster. Or am I underestimating how dark some scenes can get ?]
In the short term, you can try the following: Add more black test points using "targen -B". Try adding a lot more - 20, 50, 100, depending on your chart size. This not only gives you more of an average value, it increases the weighting of that point in the profile fitting.
Try different models, shaper+matrix, Lab cLut, XYZ cLut. One of these may better fit your particular display than the others.
In the long term I'm planning on adding a "dynamic range" factor to targen and colprof, that allows emphasizing the accuracy of the darker regions of the device behavior, with an aim to improving accuracy there.
Another possibility would be for me to add another option to collink analogous to the -f/F flags, that forces the black point. I'm reluctant to do this because the whole purpose is to not make assumptions about the device behavior, but to measure it. Forcing the black to map device 0,0,0 -> 0,0,0 makes the assumption that device 0,0,0 really is the device black point. I've seen a number of examples where this isn't true (one is right in front of me at the moment!) and a calibration/profiling systems that ignores this possibility is not doing its job. The consequences of using such a flag wrongly would be to clip the blacks and loose shadow detail. Implementation is complicated/problematic too when there are input and output one dimensional curves.