November 2, 2020

Host: Michael Dolan

Attendees:

  • [X] Mark Boorer (TSC) - Industrial Light & Magic
  • [X] Mei Chu (TSC) - Sony Pictures Imageworks
  • [X] Sean Cooper (TSC ACES TAC Rep) - DNEG
  • [X] Michael Dolan (TSC Chair) - Epic Games
  • [X] Patrick Hodoul (TSC) - Autodesk
  • [ ] John Mertic - Academy Software Foundation / Linux Foundation
  • [ ] Carol Payne (TSC) - Netflix
  • [X] Mark Titchener (TSC) - Foundry
  • [ ] Carl Rand (TSC) - Weta Digital
  • [X] Doug Walker (TSC Chief Architect) - Autodesk
  • [X] Kevin Wheatley (TSC) - Framestore
  • [X] Bernard Lefebvre - Autodesk
  • [X] Matthias Scharfenberg - Industrial Light & Magic
  • [X] J Schulte - Industrial Light & Magic

Apologies: Carol Payne

OCIO TSC Meeting Notes

  • ILM’s v2 questions

    • Doc: https://docs.google.com/document/d/11obAVucrsJ3e5HMuR40GD8fLUGAwwfbWD_jU5xId07Y
    • Mark B: Feedback on:
      • Dynamic and grading Transforms:
        • What these are trying to represent as building blocks in OCIO. Who is the target user? Useful for color sub-tasks, but not necessarily a core goal of OCIO.
        • See conversation below...
      • LogCameraTransform:
        • Parameters on transform don't coincide nicely with camera whitepapers or reference transforms. We understand the translation but a bit of a closed box to those authoring configs. Future of this transform - will additional curve characteristics be needed?
      • GetEnvVariables:
        • New get/set env vars. Do they need to be in OCIO API?
      • etc.

  • Dynamic and grading Transform discussion:

    • Mark B: For wider world, why are our attributes for transforms different from other implementations? Might impose worldview on grading which doesn't reflect industry.
    • Doug: A lot of discussion about this. Motivation explained in GH issue. Not intended to change what OCIO is used for. OCIO already has ASC CDL and Lut3D transforms, which are used for looks typically. Felt there was a gap between those extremes. Felt need for other transforms to fit between those and have advantages of both. Wasn't to change OCIO or its usage. One thing at meeting that raised a lot of discussion was not about transform themselves, but the dynamic parameters that were part of them. To modify transforms with dynamic parameters. Main reason for this is to make OCIO more usable in GPU scenarios. Without this, every change needs re-compile of shader, which was a major blocker for some GPU application usage.
    • Mark B: Main place we differ is nobody is authoring looks in OCIO. Being done in other apps. OCIO responsible for using them only. Grading transforms don't have file equivalent, that someone might want to implement. Not on same parallel in our view. CDLTransform could fall into this too, since FileTransform can handle a CDL. Custom shader code can be glued into shader.
    • Kevin: We would love to have done dynamic adjustments in single step. We're trying to do this in our playback/review tool. Much cleaner than jumping back and forth. Would be nice to be able to plug something in. Would also be nice if there was an agreed upon spec of what these grading controls meant. Do like being able to do real time adjustment of transforms, since we do that.
    • Mark B: DCC authors could need this, but not config authors. Don't see a config file equivalent to some of these knobs. Would love all of this as a DCC author, since it simplifies implementation, but not a lot of that problem in OCIO. Makes OCIO more complex and more uncertain of use case.
    • Kevin: Had not made strict limitation on where boundary should lie. Without grading transforms, dynamic parameters allow you to do thing like control where an exposure change is happening. Mostly temporary adjustments. Some of the grading transforms allow me to do things where I have previously had to bake LUT, so idea of parameterized methods is useful. I'm in the middle somewhere. Think grading transform naming is wrong and misleading.
    • Mark B: Current impl with DisplayTransform controlling where pieces are dynamic had idea of limiting (gamma, exposure) adjustments. Maybe we need more display transforms which ensure the knobs are in the proper place. Do like building block transforms to replace need for LUTs.
    • Kevin: Yes, low level ops to support this. Isn't it up to config author to choose where to express the dynamic parameters? The way these two features come together is what creates a risky area.
    • Doug: Yes, there is a specific tag to make things dynamic.
    • Mark B: New building blocks could be confusing. Grading transforms don't have file to back them up.
    • Doug: CTF format could be used.
    • Kevin: Concern with these transforms is they are not standardized. Not a big deal having additional transformations.
    • J: BLG is a black box, but fills a need in industry. CDL doesn't do that. We like fixed pipeline, and limited input to what can be changed. CDL has been building block for this. Who makes and communicates these decisions? How do we know what the right version for a parameter number is? Does that fit into OCIO core concepts?
    • Mark B: Would expect to being given CTF rather than a grading transforms with dynamic parameters set.
    • Doug: Wouldn't be handed config with dynamic parameters and expect them to edit it. Understand the concerns, but don't see how grading transforms are different from CDL in that regard.
    • Mark B: The workflows it introduces are a big shift in the tool. If someone adjusts in their config and send the config without those numbers, there's an extra bit to account for.
    • Kevin: If I got a config file with parameters that would be helpful. At moment we get CDLs and stuff that should match, but they don't sometimes.
    • J: The variability of these controls is something to help facilitate re-baking LUTs. There's value in this, but our concern is this will create dynamic version of this OCIO experience that is not intended. Can forsee a future that used builtin grading tool that uses these transforms and export that to handoff to studios, but would be a nightmare.
    • Mark B: Or combining grading transforms in row. What will that mean?
    • Kevin: Nothing stopping onset grading from baking out CLF/CTF files, so not that different.
    • J: Not sure if fix for these problems should exist in OCIO. Concern from supervising standpoint that we'll have multiple versions of a color pipeline, since it's no longer static. If Foundry opens up these tools to artists for example, that would have big effect. Not what OCIO foundationally supported.
    • Doug: Distinction want to emphasize is that Nuke right now allows editing CDL, but there is no way to insert that back into a config and write out a new one. Would be same thing with grading transforms. Someone could use this and save it, but would they be allowed to edit a config?
    • J: Don't want people affecting the state of color. Shouldn't allow means to affect that change, since that's not something we would do in our workflow. Some industries may need this though.
    • Mark B: Concern is that they can be put in config. Perhaps some parts of OCIO are for DCC tools and others for rigid color pipelines. Not sure how you can merge those things.
    • Doug: What's the concern about transforms vs non?
    • MArk: Transforms can be put in configs, whereas tools are for display.
    • Doug: Use case was for these to be put in configs, with advantages over CDL.
    • Michael: The use cases I have seen are for supes to create CDLs for looks in a review session, which can then be passed off and used in shots or in the config. The implementation is specific to a custom tool, but not something that would be exposed in DCCs. Not needing to rebuild shader interactively a big gain for this workflow.
    • Doug: Not a requirement for DCCs to make these editable. Don't see that happening. Not likely DCC is going to browse a config, edit it, and write it back out.
    • Mark B: Not concerned about writing back out; more concerned about artist changing stuff in display pipeline.
    • Doug: Would be surprised if DCC allowed you to edit looks.
    • Mark B: If you're a DCC author, there's a different track, but that problem is unrelated to config author and those maintaining transform hierarchy.
    • Sean: your thought is then that OCIO should only take image to certain point? If review tool did want dynamic parameters, it handles it on its own and then hands off to OCIO. Seems like more room for error. Nice to handle it in OCIO where transform is consistent. I see both sides points, but final control of pixels is a bit exaggerated. Every DCC has exposure and gamma slider. No concern that when I tweak those I am tweaking the config. Imagining an OCIO display with two views, one with the canonical look, and another called "editable". You would have tw switch between them, allowing a supe to make creative decisions when needed, but would require explicit switching between display chains. Don't see any operation which would require editing the original config.
    • Mark B: Not about editing the config; about expressing changes made when editing.
    • Sean: CDL-centric workflow for example. Would need to bake the changes into a file (CTF) which could be referenced. Hard though since you can't switch between dynamic and file transform. Like that this gives more capability than what a CDL can, but am safe in knowledge that decisions can be reproduced.
    • Mark B: New file format can be designed by ACES to handle additional changes. Just problem with live aspect.
    • Sean: Natural extension of DisplayTransform. Causing more issues since can be placed anywhere in config.
    • Doug: The dynamic aspect is what seems to be most troubling. See that as an implementation detail. There's nothing preventing an application from editing parameters. The dynamic parameter is only there to allow doing it interactively with GPU. Could bake out new transform continually. Whether something can be edited is separate from whether developer uses these tools.
    • Mark B: A DCC author will implement it at some point.
    • Doug: Most tools already have CDL transform.
    • Mark B: But none allow you to edit it live.
    • Doug: Dynamic feature there as a convenience. There because of exposure contrast in viewing pipeline. That tag is there to say doing exposure at front and gamma at back. Doesn't need to be there for grading parameters.
    • Kevin: Can do dynamic functionality right now with search paths, etc. We abuse that for CDL selection, an explicit choice.
    • Doug: Thanks for the feedback. We have response for all of these. In terms of written response, would you like answers in this doc, or in separate doc?
    • J: Separate doc is preferred.
    • Michael: Will be good to share these topics and conversation with wider community too, for additional input. We can continue this discussion after we all have time to think through the concerns raised.
    • Update: After the meeting, Doug provided a written response to the ILM questions: https://docs.google.com/document/d/1HeA9fJvrxAVoUQ3M1p8K3gAgd7cLkHUmChCoPfmyZvQ/edit?usp=sharing

  • Items for next TSC meeting agenda:

    • Overview of new PRs (1183, 1184, 1189)
      • #1183 - Add operator << for FormatMetadata
      • #1184 - Description in config files
      • #1189 - Named color transforms
    • CUDA support
  • No labels