Presentation

<
Towards automated corrections in video-driven animation transfer
SessionML in Production
DescriptionWe improve Digital Domain's video-driven animation transfer technique by introducing automatic corrections as a post-process optimization. We minimize the difference between our face swap model output (extended for light invariance) and predicted animation parameters in a differentiable pipeline. Our method is now being integrated into Digital Domain's facial capture workflow.
Contributors
Jose Serra
Digital Domain
Lucio Moser
Digital Domain
Event Type
Talk
TimeThursday, 14 August 20253:06pm - 3:28pm PDT
LocationWest Building, Rooms 109-110
Session TimeThursday, 14 August 20252:00pm - 3:30pm PDT
LocationWest Building, Rooms 109-110
Interest Areas
Production & Animation
Research & Education
Recordings
Livestreamed
Recorded
Keywords
Artificial Intelligence/Machine Learning
Capture/Scanning
Computer Vision
Next PresentationNext PresentationHow We Aped It: Thousands of Shots and Character Simulations with Only a Handful of TDs
Similar Presentations
Synth2Track Editor for Efficient Match-Animation
Model See Model Do: Speech-Driven Facial Animation with Style Control
The Art of Crowds Animation