HeadGAP：基于可泛化高斯先验的少样本 3D 头像
计算机视觉与模式识别
2025-01-14 v2
摘要
本文提出一种能够从有限样本野生数据中实现高保真度和可动画鲁棒性的 3D 头像创建方法。鉴于该问题的非约束性质，融合先验知识至关重要。因此，我们提出了由先验学习与头像创建两个阶段构成的框架。先验学习阶段利用来自大规模多视角动态数据集派生的 3D 头像先验；头像创建阶段则将这些先验应用于少样本定制。我们通过利用基于高斯溶解的自编码器网络有效捕获这些先验，采用基于部件的动态建模。该方法采用身份共享编码结合个性化潜在代码，以学习高斯基本块的属性。 During the avatar creation phase, we achieve fast head avatar personalization by leveraging inversion and fine-tuning strategies. extensive experiments demonstrate that our model effectively exploits head priors and successfully generalizes them to few-shot personalization, achieving photo-realistic rendering quality, multi-view consistency, and stable animation.
引用
@article{arxiv.2408.06019,
title = {HeadGAP: Few-Shot 3D Head Avatar via Generalizable Gaussian Priors},
author = {Xiaozheng Zheng and Chao Wen and Zhaohu Li and Weiyi Zhang and Zhuo Su and Xu Chang and Yang Zhao and Zheng Lv and Xiaoyuan Zhang and Yongjie Zhang and Guidong Wang and Lan Xu},
journal= {arXiv preprint arXiv:2408.06019},
year = {2025}
}
备注
Accepted to 3DV 2025. Project page: https://headgap.github.io/