Edit I’ve updated the tool with the mixing circles. That diagram only shows the effect of the inset and the outset; it does not reflect the effects of the curve, the log mapping and the hue recovery.
On the CIE xy diagram:
O is the original colour
I the inset
L the log-mapped
H the hue-recovered
R the result.
checking Outset will display the net result of the attenuation, rotation, purity boost and rotation reversal as the final gamut triangle. The effect of attenuation and rotation is always displayed.
Very strong inset/outset also affects brightness (well, even a little inset/outset does, but it’s not that visible).
Regarding the rotation sliders, I never know which direction left/right goes to which hue (green to yellow or cyan). Would it be helpful to have a coloration in the rotation sliders? I know sigmoid doesn’t have that but the rgb primaries’ Hue sliders have these shades.
But I guess with time even my little brain will learn it
I have tested the AgX OpenCL code against the regression tests. It seems to be working ok as the diff between the CPU & OpenCL is comparable to what we have for other modules. That is, the diff is between 10k to 20k pixels IIRC.
I found this strange bug where after using the white relative exposure picker after bumping exposure, a blue cast is produced that seems to stick even after disabling agx. Maybe the picker gets tied to somewhere else? It’s rather strange.
PS: Sorry for the ambient music, forgot to disable audio input in obs
I’ve noticed this with filmic in non-agx-versions of darktable as well. This might not be due to agx but something else. But I have not been able to reproduce the issue reliably enough yet.
If you look at the histogram in vectorscope mode, the effect of AgX rotation can be seen: it is the same as on the vectorscope as if you were standing behind the color and looking toward the center. So, rotate blue right and it goes toward magenta; left: cyan. Red, right: orange. Red, left: magenta. Green, right: cyan; left yellow.
I’ll leave a comment