Attendance:

Discussion:

• Nick: Working on making EXR optionally available in the USD tool chain

•  

  • Native exr importer for USD instead of OIIO

  • Using the C API

  • “Mystery challenge” to find how to load images.

  • Performance is excellent!

  • Benchmarking against tinyexr, OpenEXRCore is significantly better

  • Tinyexr errors on 80% of images in openexr-images

  • Openexr.c #include’s all files from OpenEXRCore, plus some high-level API.

  • Compiles in 14ms, without CMake

  • Larry: Unity builds in CMake: it has a feature that combines all that stuff. But only for a new CMake, 3.17

  • Nick: I bypass all the people that might object to what I want to do. I want to make it compatible with miniz, zlib alternatives. 

  • Larry: There’s also libdeflate, an alternative that is faster.

  • Rod: what’s gdeflate is based on

  • Larry: The reason I’ve been asking slack about the multipart problem, I thought I was doing something wrong. I’m pretty sure it’s a bug in Nuke.

  • The way the C core is structured, you can create parts of arbitrary size.

  • You can even change the channels encoded per chunk.

  • Subsequent chunks can have different channels. The file format doesn’t say that each chunk header has to match the previous chunk header

  • Larry: the C++ API has always bothered me, the branch where you find what kind of file it is (tiled, scanline, deep) then do separate things

  • Nick: I’m being surprised at how hard it is to use the core library, but not by fragility, behavior, etc. 

  • Larry: that’s because it’s low level

  • Cary: Could easily wrap it in python.

  • Larry: Yes, users of python probably want a single call to read pixels into numpy array

  • Nick: Hydra wants to load everything all at once.

  • Nick: It’s another crazy idea inside Pixar: if exr files are allowable as leaf nodes in a USD graph, then we can pull the camera data from the image files, so full plate info in the scene graph.

• Joseph:

•  

  • SMPTE group doing camera/lens metadata appears to be converging.

  • I propose to look at what’s come out of there, compare to what I did before, and prepare a new PR.

  • We cleaned up to reorder and group attributes.

  • I propose an initial PR that changes no methods but reorder the attributes in the groups. The subsequent PR’s won’t look so different.

  • Then, are we clearly explaining exactly what we mean

  • SMPTE declined (the people behind the recent revisions of ACES container) to add or change anything that wasn’t related to color.

  • There’s a ton of useful stuff that no one sees because they’re not in the implementation and behind a SMPTE paywall.

  • George Joblove got tagged with a bunch of Sci-Tech responsibilities