October 26, 2021

Host: Michael Dolan

Attendees: 

  • Doug Walker (TSC Chief Architect) - Autodesk

  • Kevin Wheatley (TSC) - Framestore

  • Michael Dolan (TSC Chair) - Epic Games

  • Joseph Goldstone - ARRI
  • Thomas Mansencal - Weta Digital

Apologies:

  • Carol Payne

OCIO Config Working Group Meeting Notes

  • Camera vendor contribution proposal feedback:
    • Joseph: ARRI IDTs should be expressible as CLF, and none of it is proprietary. If someone has an example for another camera, I could do them for the current camera line. ST2115 gives a general purpose, heavily parameterized log equation. Includes 4 examples. Gives general equation with vendor set of parameters. Vendors expressed their transforms in those parameters. If CLF can implement that, many transforms would be available already.
    • Doug: CLF already has CameraLog, with equation that can handle those, so we're hoping to use that. Has optimized code for CPU and GPU in OCIO. ST2115 algorithm is similar, but has parameter for high end, which CLF doesn't have.
    • Joseph: Would like to see vendor software picking the right LogC curve (EI). If we contributed LogC IDTs in CLF format, could see if vendors would consider using the correct one automatically. 
    • Thomas: Is there an official paper for the hermite splines used for EI above 1600?
    • Joseph: No. May be due to knowing users needing to linearize wouldn't use EI 1600 and above. It would have made the whitepaper more complicated, and likely not be used.
    • Doug: Other thing we've been discussing for studio config is to try and reduce the number of IDTs. Canon has a large number of variations for example. Considering asking camera vendors to recommend a subset.
    • Michael: Have been planning to put only LogC EI 800 in there initially, but could add more. Config is generated from spreadsheet, so is extensible.
    • Joseph: Could you put them all in spreadsheet and comment them out?
    • Thomas: Yes. If we just had the original repository without spreadsheet, would not be obvious. CSV file is generated from spreadsheet in Google.
    • Joseph: Do you expect a facility to take the off the shelf config with fewer transforms?
    • Kevin: Multiple use cases. Some will use it as-is, others will want super accurate linearization, and then others will take the general approach and add more specific transforms if needed. Users who want the other transforms will generally know enough to add them.
    • Joseph: EI 800 is the right choice if you only have one. Most productions are using that.
    • Kevin: Tracking details like EI across a production is complex, and for users who know more specific transforms will make a difference, they can add it. In VFX we'll get a transcoded EXR, and that's where we start. Would need the information in the file to select the right curve, but software is not putting it in there, or reading it.
    • Joseph: A lot of this data is available in ACES container spec, but not in EXR implementation.
    • Michael: Any suggestions on working with camera vendors? Open to them contributing CLF, or helping with implementation from a paper.
    • Joseph: If you get a paper, always check if there is a newer version available.
    • Doug: Planning on sending this doc to the different camera vendors. Looking for any feedback.
    • Joseph: If you give an example for S-Log or V-Log, would be pretty simple to plug in the data.
    • Michael: Want to get vendors involved to contribute, and to help maintain and provide new transforms as new cameras are available, to keep configs up to date and accurate.
    • Joseph: Makes sense. When you are running with the IDT expressed this way, is it doing any interpolation, or just an accurate GPU calc?
    • Doug: No interpolation. Just an equation on CPU and GPU.
    • Joseph: So you get the benefit of CTL without its limitations.
    • Doug: Not as flexible as CTL, but goal of CLF was to strike balance between flexibility and performance. Able to do a lot in a compact way. A lot of camera vendors have adopted similar equation with different numbers. For IDT part, hoping most people can submit these compact CLFs. Easier to read and work with.
    • Joseph: Will take a look at document, and also do comparisons with CTL. Is there a Metal implementation?
    • Doug: Apple made PR with Metal implementation recently.
  • Proposed transform ID revision in document:


  • No labels