UniHair : Towards Unified 3D Hair Reconstruction from
Single-View Portraits

SIGGRAPH Asia 2024

Yujian Zheng^{1, 2} Yuda Qiu² Leyang Jin²
Chongyang Ma³ Haibin Huang³ Di Zhang³ Pengfei Wan³ Xiaoguang Han^{2, 1}*
¹FNii, CUHKSZ    ²SSE, CUHKSZ    ³Kuaishou Technology
*Corresponding author

[Paper]      [Video]      [Code]      [DataSet]

We proposed a novel strategy which enables single-view 3D hair reconstruction with rich textures and complicated shapes, working for braided and un-braided hairstyles through a unified pipeline. From left to right: input image, reconstructed 3D hair in 3D Gaussian representation and its rendered images from multiple views. Note that, the underlying virtual avatar is just for visualization.

Abstract

Single-view 3D hair reconstruction is challenging, due to the wide range of shape variations among diverse hairstyles. Current state-of-the-art methods are specialized in recovering un-braided 3D hairs and often take braided styles as their failure cases, because of the inherent difficulty to define priors for complex hairstyles, whether rule-based or data-based. We propose a novel strategy to enable single-view 3D reconstruction for general hair types via a unified pipeline. To achieve this, we first collect a large-scale synthetic multi-view hair dataset SynMvHair with diverse 3D hair in both braided and un-braided styles, and learn two diffusion priors specialized on hair. Then we optimize 3D Gaussian-based hair from the priors with two specially designed modules, i.e. view-wise and pixel-wise Gaussian refinement. Our experiments demonstrate that reconstructing braided and un-braided 3D hair from single-view images via a unified approach is possible and our method achieves the state-of-the-art performance in recovering complex hairstyles. It is worth to mention that our method shows good generalization ability to real images, although it learns hair priors from synthetic data.

Overview

Results

Comparisons

Hairstyle Customization for 3D Avatars

Reference

@article{zheng2024unihair,
  title={Towards Unified 3D Hair Reconstruction from Single-View Portraits},
  author={Zheng, Yujian and Qiu, Yuda and Jin, Leyang and Ma, Chongyang and Huang, Haibin and Zhang, Di and Wan,Pengfei and Han, Xiaoguang},
  journal={arXiv preprint arXiv:2409.16863},
  year={2024}
}

Acknowledgements:
Website template is borrowed from gan_steerability.