UI Binding Transfer for Bone-driven Facial Rigs

Abstract

We propose an automatic method to transfer the UI binding from a rigged model to a new target mesh. We use feed forward neural net-works to find the mapping functions between bones and controllers of source model. The learned mapping networks then become the initial weights of an auto-encoder. Then the auto-encoder is retrained using target controllers-bones pairs obtained by the mesh transfer and bones decoupling method. Our system only requires neutral expression of the target person but allows artists to customize other basic expressions, and is evaluated by the semantic reproducibility of basic expressions and the semantic similarity.