Recap:
- @matthias.scharfenber noted that the nuke script in the research folder to calculate the distance limits/threshold is still the old CC24 values - but the difference is minuscule so should not really affect anything.
- @sdyer has the update removing the clamp on the AP0 to ACEScg CTL in his fork of the repo, and has filed a bug report so it’s tracked and will figure out how to get to released
- @KevinJW - for the distance calculations, may be helpful to instead of looking at the boundaries - look at the intersections, and use that to calculate the inverse
- Around the current default of 1.2 for the power compression value: @jedsmith notes that 1.0 is reinhard, and values higher than that tend towards C2 continuity. So 1.2 was a default picked subjectively.
- Half-float quantization could be used to start, and subjective tests from there. @nick suggested a high/medium/low “aggressiveness” exposed to the user.
- Blue-screen test shoots:
- Highly saturated object - in and out of focus
- Glass bottle
- Reflective things: safety vest/jacket, bike reflector, tin foil
- Light trap
- Emissive light (lens flares?)
- @jedsmith asked whether using the max of the vendor gamuts makes sense, instead of possibly taking an average of the out of gamut values from the test imagery (reasoning being those examples were put up as extreme examples). @matthias.scharfenber pointed out that it gets slightly subjective, and also work might be done in the camera encoding primaries and therefore might be needed. Consensus is to continue will camera encoding primaries in testing and go from there.
- Can the algorithm be encapsulated in a 3d LUT? @matthias.scharfenber noted that the key is in the shaper function. You need to encapsulate the negative values. Possible to use a reflected ACEScct curve.