Recap:
-
The majority of the meeting was @jedsmith giving a nicely broken down demo of the components of his algorithm - he will be posting the script for reference, but please see the recording for full details
-
There was a lot of talk around what ‘success’ looks like: hue shifts? Fidelity of texture (grain)? Slope of compression curve for a given bit depth?
- On hue shift:
- @LarsB voiced concerns about the hue shifts caused by the non-uniform compression between the RGB channels
- An argument against that was that within the “zone of trust” hue would be preserved.
- Lars presented Munsell hue line plots in various color spaces to demonstrate the lack of hue linearity in all of them
- On bit depth:
- We did not really discuss it on the call, but we are in floating point scene linear in ACES. This might come into play in the output transform group. Open to opinions on this in the next meeting.
- On texture/grain:
- We did not really get to this, looking for tests to demonstrate the workability of the current algorithm as it pertains to grain
- On hue shift:
-
Should we be limiting to a gamut for reasonable values? (i.e. e-gamut) how much does it really help?
-
@Thomas_Mansencal brought up again the question of whether we should be compressing to a gamut of physically realisable colors. BT.2020 vs AP1?
-
When iterating on a particular algorithm we should start using version numbers to more easily identify which iteration is being discussed in forum posts