September 8, 2020

Host: Michael Dolan

Attendees:

• [X] Mark Boorer (TSC) - Industrial Light & Magic
• [] Sean Cooper (TSC ACES TAC Rep) - DNEG
• [X] Michael Dolan (TSC Chair) - Epic Games
• [X] Carol Payne (TSC) - Netflix
• [X] Doug Walker (TSC Chief Architect) - Autodesk
• [X] Kevin Wheatley (TSC) - Framestore
• [X] Matthias Scharfenberg - Industrial Light & Magic
• [X] Dennis Adams - Sony
• [X] Thomas Mansencal - Weta Digital

Apologies:

• Sean Cooper

OCIO Configs Working Group Meeting Notes

• PR #5 - “aces-dev” Reference Config Generator:

  • Thomas: In progress. Big missing piece is transforms, which are not mapped yet, but color spaces should all be there.
  • Michael: Question about "undefined" display. Can that be named?
  • Thomas: Trying to massage CTL to get names. aces-dev not ready to be ingested like this. Transform IDs are busted. Display is "undefined" since it contains transforms that are not categorized. Scott will classify these.
  • Michael: HDR most likely?
  • Doug: Ideally classified as P3-D65, ST2084, etc.
  • Thomas: BuiltinTransform next piece. Then talk about what to do for displays, display reference space, etc. Doing some string massaging, which prevents matching old ACES config. Should keep names consistent where possible.
  • Doug: There is an issue in OCIO GH talking about improved way to deal with aliases. Current ACES config has many aliases for including in file paths, but it’s a maintenance problem since the color spaces are duplicated. Could add an alias feature to color space in core OCIO.
  • Michael: Have written code to alias by duplicating color spaces in the past. Alias feature would be useful.
  • Doug: Without fallback mechanism, will create unhappy users.
  • Michael: Would create clutter trying to be backwards compatible without that.
  • Thomas: Maybe we can create a shim, or document/tool for migration. Otherwise we're building on dirty foundation.
  • Mark: Also missing the new color space name formatting. Had discussed move to transfer function and gamut convention. If we were to move forward, would be good to make ground truth adhere to that first.
  • Thomas: Tried to be consistent, but we can change the way things are named.
  • Michael: Tricky because would require scripts to edit DCC scene files. Maybe just provide data for mapping in implemented pipeline code.
  • Thomas: It won't just be names. There is stuff that needs to go away to clean up config. Mapping doesn't cost much, can have it.
  • Kevin: What has been generated makes sense for core ACES, since it follows CTL. The more encompassing config could follow the old naming to help with migration. Could be fit into generator. Color space names issues should be fixed on ACES side, if users want to adhere to ACES.
  • Thomas: Problem is if we allow users to stick to old naming it will prevent migration. Maybe provide method, but not direct support.
  • Doug: If we had ability to define alias with color space, a DCC would use the official name, so if someone opens something that uses the old name, the DCC could do conversion automatically through alias system.
  • Mark: Don’t think it will matter much. Not great when DCC writes color space name to file. Can use roles now to accomplish this.
  • Matthias: Roles would perpetuate the old names.
  • Mark: Don’t think DCC needs to be considered as long as names resolve to the correct color space.
  • Doug: Our DCCs do need to store color space names to save color management behavior. Could be manageable to convert names though.
  • Matthias: Would this lookup be provided as a separate file? Or would this be provided as part of a color space definition? And DCC would see that it’s an alias and update the string value to the actual color space?
  • Doug: The proposal on GH is just to add alias attribute with list of strings. Haven't figured out details, but an API call could be added to check if a name is an alias or not, to get the canonical name.
  • Mark: That would help the DCC but not the user. If the old name had a different family for example, would that require family aliases, etc?
  • Matthias: Trying to avoid users perpetuating old names. Can put that in documentation somewhere.
  • Mark: How to handle the months during transition. Is it worth the effort, or do we encourage starting fresh?
  • Doug: Wasn’t clear to me that aliases need to show up in UI at all. Internal convention.
  • All: Agree that aliases don't need to show up in UI.
  • Mark: It’s mostly so that a DCC that saved with the old name can continue to work with the new name.
  • Carol: People will want to know what the changed color space names mean.
  • Mark: Only way to describe is with description string, or in name.
  • Doug: Was thinking most of these changes would be fairly clear, subtle renames. Different, but similar. Names are informed from string. Previous aliases were for convenience for filename and commandline tools.
  • Carol: Yes, would be cool to have that as alias, but not color space name.
  • Mark: Think all color space names should be clear, short names, and put more data in description.
  • Kevin: Also issue of AMF files referring to color spaces by name from ACES.
  • Thomas: We could also store full ACES transform ID as alias. Ground truth.
  • Mark: That’s AMF consideration though. Nothing to do with us.
  • Carol: Has to do with us if we need to relate AMF to OCIO config.
  • Mark: There’s been little coordination from AMF team about how that information is supposed to flow into VFX world. Will be responsibility of AMF reading tool and authors to handle these details.
  • Carol: If we’re designing an ACES reference config, AMF is part of ACES, so we need to consider how the whole ACES pipeline works together.
  • Kevin: Agree with Mark that we won't use this internally, but we will have to deal with it with vendors who we need to hand files off to.
  • Mark: We’re supposed to represent ACES in a way that works well for OCIO, which may not be perfectly inline with how ACES community names and structures their files. Much more concerned with how this config will work as OCIO config. Less about AMF and Academy structure decisions.
  • Thomas: It is an ACES config though.
  • Carol: Be better if we worked together rather than working in silo. We can work with ACES to make a better OCIO config.
  • Doug: at the end of day, naming discussions are contentious. The transform ID strings are a bit missed up, but Scott will probably fix that. This may be a logo requirement too. To be product partner we’ll need to follow logo guidelines. The name that shows up in UI is under their purview. Open to our participation, but not realistic to sway their naming based on our limitations (i.e. name in filename).
  • Carol: Think the alias solution is a good solution. Pretty elegant and don't think people would have a lot of issue with it.
  • Thomas. Yeah, at end of the day its a UI issue. DCC can choose to switch the display of color space on their end.
  • Doug: From a DCC perspective, we’re trying to let the config author decide that. At the end of the day the config should define the names in menus, but maybe DCC could add option to see aliases for canonical name. More a DCC specific choice based on requests from user base.
  • Mark: If we make blanket statement that alias never shows up in DCC, then they have to be entered as separate color spaces with duplicate entries in config.
  • Doug: Would be possible to get alias reference from API.
  • Mark: Think it would be better for aliases to be only for backwards compatibility.
  • Doug: Yes agree. If you need something to show up in menu, should be added as duplicate color space.
  • Carol: From a user perspective, the general user may not understand that the canonical and alias names are the same and can be used interchangeably. That’s a concern we hear a lot.
  • Matthias: Could aliases show up in description string in tooltip in DCC?
  • Dennis: Is there a recommendation on creating hierarchy based on separators?
  • Doug: Family can be used for that and some code in OCIO helpers will help do that. Now a 'family_separator' property to help standardize.

• BuiltinTransform skeleton, implementations:

  • Michael: Now that the config skeleton is created, will need to create skeletons for BuiltinTransform in core OCIO. Thoughts on how to proceed?
  • Doug: Builtin transform accessed with a name string. Ideally with this mechanism in place to pull names from ACES, would be cool if those strings could match up in core OCIO to avoid manual interventions.
  • Thomas: All the strings you see here are programmatically generated from CTL transform ID. Can be used to call transform. Could use CTL transform ID. Will try adding transforms next week to see how config generates.
  • Michael: Wait for config update to touch C++?
  • Thomas: Yes, in next two weeks will find some time.
  • Michael: Will all builtin transforms use existing ops? Is anything missing algorithm-wise needed to support ACES?
  • Doug: Some, like PQ will be represented with LUT, but can be generated internally.
  • Thomas: Maybe the current generator could generate core skeleton from aces-dev. Instead of manually going through and adding everything. Maybe could generate the files. Just a suggestion.
  • Mark: Think we want to keep those things separate. Be human added.
  • Thomas: Could be a lot of files.
  • Mark: Want someone to vet them.
  • Thomas: Agree. Generation code could just generate them, and then review that code manually. Think will need a discreet file per edge in the graph, and there are a ton of edges in graph. If you want to reference implementation, we need a transform for each edge.
  • Mark: Your current PR is more than we would consider in the library. Current config has camera transforms.
  • Thomas: Yes, IDTs can be removed. Let’s talk just about output transforms. Think we need a builtin transform for each discreet ODT.
  • Doug: For output transforms, it’s a bit of a separate case. Will make proposal of how to use display reference space and view transforms to reduce the builtin transform requirement.
  • Kevin: Could write a parser for CTL to generate OCIO properly. Detect matrix operator, etc.
  • Mark: Working on CTL parser for OpenColorMath, which was hoping could be used here. Then could reference CTL directly.
  • Michael: Can always update implementation later.
  • Doug: Want something in near term to help people use new OCIO. Going to update config writing to not write default values, which will reduce config size.
  • Mark: If no equality groups, entire config is a no-op. That will need to be addressed.