Official repository for the paper

GeoDiff4D: Geometry-Aware Diffusion for 4D Head Avatar Reconstruction

Chao Xu^1,†, Xiaochen Zhao¹, Xiang Deng¹, Jingxiang Sun¹, Donglin Di² Zhuo Su^1,‡ Yebin Liu^1，§

¹Tsinghua University, ²Li Auto

^† Work done during internship. ^‡ Project leader. ^§ Corresponding author.

[Paper] [Project Page] [Citation]

The Video Generation Model is based on X-NeMo. The 4D Gaussian avatar code is based on CAP4D. Special thanks to the authors for making their code public !

Related work:

• GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians
• FlowFace: 3D Face Tracking from 2D Video through Iterative Dense UV to Image Flow
• StableDiffusion: High-Resolution Image Synthesis with Latent Diffusion Models
• Pixel3DMM: Versatile Screen-Space Priors for Single-Image 3D Face Reconstruction
• VHAP: Versatile Head Alignment with Adaptive Appearance Priors
• DAViD: Large-Scale Synthetic Dataset of Human Body
• NeRSemble: A Large-Scale Multi-View Video Dataset of Facial Performances
• RenderMe-360: High-Fidelity Multi-View Human Head Dataset

Awesome concurrent work:

• CAP4D: High-Resolution Multi-View Humans from a Single Image
• MVP4D: 360-degree 4D Avatars from a Single Image
• Avat3r: Large Animatable Gaussian Reconstruction Model for High-fidelity 3D Head Avatars

May these excellent works offer help to those in need.

@misc{xu2026geodiff4dgeometryawarediffusion4d,
      title={GeoDiff4D: Geometry-Aware Diffusion for 4D Head Avatar Reconstruction}, 
      author={Chao Xu and Xiaochen Zhao and Xiang Deng and Jingxiang Sun and Zhuo Su and Donglin Di and Yebin Liu},
      year={2026},
      eprint={2602.24161},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.24161}, 
}