RegHead: Non-Humanoid Head Blendshapes via Feed-Forward Registration

2026-07-13

AI Summary

This work addresses the challenges of constructing semantic blendshapes for non-humanoid heads, which include scarce supervision data, lack of correspondences in 4D assets, and highly localized facial motions. The authors propose the first efficient blendshape construction framework tailored for non-humanoid avatars: starting from a small set of artist-provided expressions, they fine-tune image editing models to synthesize a large-scale dataset of non-humanoid characters; they introduce dense stochastic anchors to represent local deformations and design a feedforward mesh registration network that rapidly generates semantically consistent blendshape bases with vertex correspondence from unaligned expression meshes. The method achieves superior fidelity compared to existing approaches, offers inference speeds orders of magnitude faster than optimization-based techniques, and enables real-time, accurate retargeting of human facial motion capture signals onto non-humanoid characters.

Abstract

We present RegHead, a framework for constructing semantic blendshape sets for animatable non-humanoid head avatars. With a fixed expression vocabulary, semantic blendshapes provide a low-dimensional and interpretable animation interface and support cross-identity retargeting. Building such blendshape sets remains expensive because (i) expression-consistent supervision is scarce, (ii) generated 4D assets typically lack correspondence, and (iii) facial motion is highly localized. We propose (1) a large-scale dataset of non-humanoid identities paired with a shared expression vocabulary, obtained by expanding a small artist-rigged library via fine-tuned image editing; (2) a dense stochastic anchor motion representation tailored to localized facial deformations; and (3) a fast feed-forward registration model that converts unregistered expression meshes into a corresponded blendshape basis by predicting anchor-based deformations from the neutral shape. Experiments show that our approach produces higher-fidelity expression meshes than baselines, while running orders of magnitude faster than optimization. We further demonstrate real-time retargeting from human face tracking signals to non-humanoid characters, capturing both head pose and localized facial motions. Our project page is available at https://snap-research.github.io/RegHead/.

Problem

• non-humanoid avatars
• semantic blendshapes
• expression correspondence
• facial animation
• 4D asset alignment

Innovation

• blendshapes
• non-humanoid avatars
• feed-forward registration
• facial animation
• expression retargeting