Face, Body, Interaction: The Future of Digital Humans?
Do we still need explicit representations for faces and bodies? As diffusion and vision-language models generate ever more convincing video of people, the temptation is to declare the parameterization era over. Pixels are the representation we have the most data for, but not necessarily the right abstraction for real-time, interactive, and controllable characters. In this keynote I will argue that interaction — not fidelity — is the key requirement for impactful digital human applications, and explore what that implies for implicit vs. explicit approaches.
- Keynote
Speaker
Iain Matthews
Courtesy Faculty, Robotics Institute, Carnegie Mellon University; Honorary Professor, School of Computer Science, University of East Anglia; Former Director, Research Science, Epic Games
Iain Matthews has spent two decades building systems at the intersection of computer vision, graphics, and machine learning for digital humans. He most recently led a Research team at Epic Games, shipping real-time ML systems for facial animation, body deformation, and character pipelines in Unreal Engine and MetaHuman. Earlier, he created the facial motion capture system used on Avatar and Tintin at Weta Digital, for which he received a Scientific and Engineering Academy Award in 2017. Iain hol... read more