Neural Facial Deformation Transfer

How to read this

Category

Method: neural facial expression / deformation transfer

Contributions

• Neural Facial Deformation Transfer (NFDT), a data-driven method that transfers facial expressions from a template character to new targets given only the target's neutral shape.
• A data generation strategy that automatically builds a large training set of paired template/target shapes in the same expression.
• A topology-agnostic decoder-only transformer (adapted from the Shape Transformer) that transfers without a rig inversion step and generalizes across mesh topologies and to humanoid creatures.

Context

Extends facial performance retargeting beyond the authors' prior local anatomically-constrained retargeting, moving to a topology-agnostic, rig-inversion-free neural transfer.Builds on: Local Anatomically-Constrained Facial Performance Retargeting

Correctness

Reported to outperform prior expression-transfer methods in quantitative evaluations and a user study; the key dependency is the automatically generated paired-shape data, so a reader should keep in mind transfer quality is bounded by that synthetic data and by how well targets resemble the training distribution.

Clarity

Accessible problem framing; a first pass conveys the transfer setup, do a second pass for the transformer architecture and the data-generation strategy.

How to read it

Focus on the neutral-shape-only input, the rig-inversion-free design, and the data-generation strategy; second pass the Shape Transformer adaptation if implementing.