Virtual Meeting: Machine Learning in Visual Effects

Hey Logik :slightly_smiling_face:

For anyone into machine learning: SMPTE Hollywood and the Visual Effects Society are cohosting a meeting exploring the impact of machine learning on visual effects next Wednesday, January 26th, at noon PT.

Panelists include:
Will Harris, Flame Family Product Manager, Autodesk
Mathieu Mazerolle, Director of Product, New Technology, Foundry
Brian Gaffney, Business Development Manager, Unity Technologies
Ryan Laney, Visual Effects Supervisor, Teus Media
Moderated by Barry Goch, Senior Editor, ColorTime

You can register for free here: https://autode.sk/3IDgeJl

3 Likes

I don’t wish to rain on the parade of this, but it would be great to see ML kick on in Flame as it seems to have been introduced and it is not no longer going the great guns that it departed the station on, without arriving at its destination. Some aspects blew my mind, in terms of the imaginative usages, and how it rippled out into Flame’s toolsets (like that fantastic idea of holding out people/moving in the new camera tracker through ML!), but the instability of generated spaces has remained (the instability of the mattes should not be the stable signature feature!) and so this really needs “finishing.” Hopefully this gets going again soon as it’s been a bit of a long-ball feature (a soccer metaphor). Still have high hopes for this, but it’s not a binary feature that hit it’s 1.0 presence.

Cheers
Tony

3 Likes

TonyRichards –

You might enjoy this link, especially the video under the “Video (Paper)” heading:
https://isl-org.github.io/PhotorealismEnhancement/

While the “photorealism enhancement” is very impressive (bonus: seeing Grand Theft Auto’s vision of Los Angeles overlaid with a German color palette is interesting!) what I was most impressed by is how stable is their process.

When theirs is split-screened with competing techniques it’s immediately obvious that their approach is superior in its temporal stability.