Saved in:
Bibliographic Details
Main Authors: Ji, Xinya, Weiss, Sebastian, Kansy, Manuel, Naruniec, Jacek, Cao, Xun, Solenthaler, Barbara, Bradley, Derek
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.13837
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917234311430144
author Ji, Xinya
Weiss, Sebastian
Kansy, Manuel
Naruniec, Jacek
Cao, Xun
Solenthaler, Barbara
Bradley, Derek
author_facet Ji, Xinya
Weiss, Sebastian
Kansy, Manuel
Naruniec, Jacek
Cao, Xun
Solenthaler, Barbara
Bradley, Derek
contents Despite recent progress in 3D Gaussian-based head avatar modeling, efficiently generating high fidelity avatars remains a challenge. Current methods typically rely on extensive multi-view capture setups or monocular videos with per-identity optimization during inference, limiting their scalability and ease of use on unseen subjects. To overcome these efficiency drawbacks, we propose FastGHA, a feed-forward method to generate high-quality Gaussian head avatars from only a few input images while supporting real-time animation. Our approach directly learns a per-pixel Gaussian representation from the input images, and aggregates multi-view information using a transformer-based encoder that fuses image features from both DINOv3 and Stable Diffusion VAE. For real-time animation, we extend the explicit Gaussian representations with per-Gaussian features and introduce a lightweight MLP-based dynamic network to predict 3D Gaussian deformations from expression codes. Furthermore, to enhance geometric smoothness of the 3D head, we employ point maps from a pre-trained large reconstruction model as geometry supervision. Experiments show that our approach significantly outperforms existing methods in both rendering quality and inference efficiency, while supporting real-time dynamic avatar animation.
format Preprint
id arxiv_https___arxiv_org_abs_2601_13837
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FastGHA: Generalized Few-Shot 3D Gaussian Head Avatars with Real-Time Animation
Ji, Xinya
Weiss, Sebastian
Kansy, Manuel
Naruniec, Jacek
Cao, Xun
Solenthaler, Barbara
Bradley, Derek
Computer Vision and Pattern Recognition
Despite recent progress in 3D Gaussian-based head avatar modeling, efficiently generating high fidelity avatars remains a challenge. Current methods typically rely on extensive multi-view capture setups or monocular videos with per-identity optimization during inference, limiting their scalability and ease of use on unseen subjects. To overcome these efficiency drawbacks, we propose FastGHA, a feed-forward method to generate high-quality Gaussian head avatars from only a few input images while supporting real-time animation. Our approach directly learns a per-pixel Gaussian representation from the input images, and aggregates multi-view information using a transformer-based encoder that fuses image features from both DINOv3 and Stable Diffusion VAE. For real-time animation, we extend the explicit Gaussian representations with per-Gaussian features and introduce a lightweight MLP-based dynamic network to predict 3D Gaussian deformations from expression codes. Furthermore, to enhance geometric smoothness of the 3D head, we employ point maps from a pre-trained large reconstruction model as geometry supervision. Experiments show that our approach significantly outperforms existing methods in both rendering quality and inference efficiency, while supporting real-time dynamic avatar animation.
title FastGHA: Generalized Few-Shot 3D Gaussian Head Avatars with Real-Time Animation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.13837