Realtime Performance-Driven Physical Simulation for Facial Animation

How to read this

Category

Method: real-time physics-based facial animation from performance capture

Contributions

• Presents the first real-time method enhancing performance-driven facial animation with physical simulation of volumetric skin, lip contacts, and sticky lips
• Introduces an efficient progressive Projective Dynamics solver to handle constantly changing constraints in real time on the CPU
• Adds a Taylor approximation and memoization scheme to speed up the SVD computations needed for volumetric skin simulation, driving blendshape characters from a simple webcam feed

Context

Builds on the authors' BlendForces dynamic facial-animation framework (Barrielle et al. 2016) and on blendshape-based capture, adding real-time physical simulation on top of a blendshapes model without requiring physical or medical data.Builds on: BlendForces: A Dynamic Framework for Facial Animation

Correctness

Relies only on a blendshapes model (no medical/physical data) and is demonstrated driving characters from webcam input; real-time CPU performance hinges on the progressive solver and SVD approximations, whose accuracy tradeoffs a reader should keep in mind.

Clarity

Technically dense in the solver sections; a first pass conveys the contributions and pipeline, a second pass is required for the Projective Dynamics and SVD-approximation details.

How to read it

First pass for what physical effects are added and the practical webcam-driven setup; second pass on the progressive Projective Dynamics solver and SVD memoization if real-time performance or the numerics matter to you.