2024-09-16 TSC

ACES 2.0 planning (library and configs)

• Working through testing round trips with the images provided by Scott - some issues with NaNs popping up on the inverse. Only affecting a few pixels per 3k image though, otherwise it looks good.

• Will be pushing some small fixes, floating point precision fixes, bound checks before sampling lookup tables

• Doug: 50 OTs in the list from AMPAS, (25 each per whitepoint) - some we’re waiting on a built-in display for, making progress there. How many are you actually able to check now?

• Remi: can check everything because I made a custom config to get around the built-in missing.

• What sort of tolerance are you using? Remi: currently using idiff and look at the stats. No specific threshold set. Can post that on the PR, including the script used to generate. Currently min difference is less than 5 decimal places, so it’s pretty good.

• Doug: so confident we have a matching version to the CTL?

• Remi: yes, confident. It’s just edge cases now.

• So the main thing for 2.4.1 will be performance, and trying to remove the cusp tables.

• Goal is to have a final PR by EOW.

• Doug will review, Zach will also review. Any other folks with cycles, we’d really appreciate a look. Thomas will try to give it a look on Windows.

• Michael will reach out to Eric Renaud-Houde at Epic to see if he can re-check the HLSL.

• Remi: wheels are broken for python 3.13, and on windows. nightly build is failing, don’t have time to take a look.

• Thomas: will take a look, we can work offline together.

• Configs - please test the ACES 1 configs - can test with OCIO 2.x, main addition for 2.4 is Apple Log

• ACES 2.0 configs will come later with PCs for the configs and then built in to 2.4.1

OCIO 2.4.0 release plan

Must release by Sept 30!

PRs:

• OCIO 2.4.0 built-in ACES 1.3 config updates

• Remi PR #1983 (ACES 2)

• Cuneyt PR #2029 (FixedFunction improvements)

• Doug PR (new built-in displays for ACES 2)

• Rosie PR #1984 (RGBA textures for Metal)

• Thomas PR #2023 (ocioview clamping)

• Remi PR #2038 (ocioconvert bit-depth)

• Any others?

Color Interop Forum Spec

• Looking to finalize spec by the end of the week

• Please give us any feedback ASAP

Dev Days

• looking for better "good first issues"

• "office hours" sign up (https://docs.google.com/spreadsheets/d/1ILMDlWbkLvxOgDn1_18BI5-eM12ozRA5aiCKf9Y3Nto/edit?usp=sharing )

Use next week for TSC meeting?

• Instead of the UX meeting? Since it’s the last bit before the release?

• Mark & Michael are fine with that.

• Hopefully there won’t be much to discuss, but just in case

HLG Built-in Display Bug

• Edge case: if you have something with very small luminance, the slope goes to inf as it approaches 0. Normally, this is fine because you’re applying it to RGB. More of an issue when it’s just used for Y in HLG. In colour and ACES, does something different if Y is 0. But this doesn’t handle super small values.

• Thomas: maybe adding a super small epsilon? Doug: that’s what OCIO currently does.

• Might be something we need to talk directly to HLG folks about, maybe Nick can make some connections at BBC.

• You don’t really run into this with 3D LUTs - really only analytic conversions.

• Also adding mirroring at 0 instead of clamping, even though it’s not in the spec, it’s what is in colour.