Originally Posted by N3W813
In the chart below you can see results the power gamma 2.2 to display is calibrated to. There should be enough room for the 3DLUT to work with to transform to a BT.1886 2.4 gamma curve. I've also included calculated BT.1886 2.4 targets from the BT1886CalcV3.xls (attached) and from the SpectraCal Calman v5 calibration software. They basically have the same targets with negligible differences. On the other hand, the actual measured grayscale with the 3DLUT in MadVR is very different from these targets, especially from 10-60%.
Looks exactly as expected if you used the "black point compensation" profile.
Since the BPC profile is lying to collink, collink thinks that the display has a zero black, and bt1886 reduces to a pure 2.4 power curve:
50% mark for 3dlut: log(21.45-.0486/110.89)/log(0.5) = 2.37, pretty close to 2.4
But I think there are a couple of basic problems with the whole idea of BT.1886 being used as per the spec. with profiles:
* Typical instruments struggle for accuracy in measuring displays with a very dark black point.
* For display with very dark blacks, the profile doesn't capture the black point accurately enough (it may be a small absolute error of 0.02 cd/m^2 which compared to white is 0.02%, but compared to the true black is a 50% error).
* BT.1886 effective gamma (the effect it has on the overall brightness of the image) is quite sensitive to the black point value.
So I think I will proceed with changing the defaults in collink to aim for an effective (default) gamma of 2.2 and make the gamma adjustment value also be effective gamma. That way the apparent brightness will not be sensitive to any issues with black point accuracy. I'll provide a different option to specify technical gamma for those who want to try pure BT.1886, etc..
The problem with "BPC off" I think is due to a mistake on my part - I was converting the display point to input space (Rec709) using a reverse lookup of the input profile that clipped to the input gamut, and if the display black point is not neutral (as is the case here), the clipping brightened up the black point a lot to get it in gamut.
I'll change the code somewhat to clip it in a different way, that doesn't brighten it up as much. If this still proves to be objectionable, then rather than try to correct the black point neutrality as much, I can try blending it to the true black point (but once again, inaccuracies in the readings and profile model near the black point can mess this up somewhat.)