Data-driven extraction and composition of secondary dynamics in facial performance capture

Our work aims to compute and characterize the difference between the captured dynamic facial performance, and a speculative quasistatic variant of the same motion should the inertial effects have been absent.