Pose-Based Facial Animation Retargeting

High-quality facial animation retargeting is crucial in digital entertainment production, enabling the transfer of animations from a source rig to a target rig. To avoid the need of manual transfer, automatic retargeting methods can be employed. However, creating universal methods remains challenging due to varying rig and topologies. We propose a novel retargeting method based on the idea that any facial expression can be decomposed into a weighted sum of Facial Action Coding System (FACS) poses. Our method assumes that these decompositions transfer across rigs with minimal need of refinement. We utilize a machine learning model to extract FACS weights from source frames. This model is trained solely on source animation data, requiring no target data or complex correspondence mappings. After a refinement step, these weights are then applied to the target rig’s FACS poses. Our technique supports seamless animation transfer across different rigs while allowing for artistic modifications by adjusting individual FACS poses. Our experiments demonstrate that this method produces high-quality retargeted animations and outperforms other retargeting methods. We explore how different design choices impact retargeting quality and showcase artistic control of asymmetries. By simplifying the workflow for rigged heads, this flexible framework enhances production efficiency while maintaining artistic control.