FHAvatar

Fast, high-fidelity reconstruction of composable face-and-hair 3D head avatars from few casual captures.

CVPR 2026 Co-author Animatable 3D Avatars
FHAvatar reconstruction, animation, and editing results

Takeaway. A fast few-shot 3D Gaussian avatar framework with composable face and hair representations, supporting real-time animation and editing.

Problem

High-fidelity animatable head avatars typically require carefully captured multi-view data or lengthy per-subject optimization, limiting practical personalization from casual observations.

Method

FHAvatar reconstructs high-fidelity, animatable 3D head avatars from only a few casual captures. It models face and hair as composable components so each region can use a representation suited to its geometry and appearance. An aggregated transformer backbone learns cross-view priors from multi-view data to support fast personalization for new identities.

Contributions

  • Composable representation: face and hair are modeled as distinct but coordinated components.
  • Fast personalization: a learned cross-view prior supports reconstruction from sparse casual observations.
  • Interactive use: the avatar supports real-time animation, hairstyle transfer, and stylized editing.
Paper on arXiv

Citation

@inproceedings{sun2026fhavatar,
  title     = {FHAvatar: Fast and High-Fidelity Reconstruction of Face-and-Hair Composable 3D Head Avatar from Few Casual Captures},
  author    = {Sun, Yujie and Cai, Zhuoqiang and Niu, Chaoyue and Chen, Jianchuan and Chen, Zhiwen and Lv, Chengfei and Wu, Fan},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year      = {2026},
  arxiv     = {2603.23345}
}

References