Abstract
Gaussian splatting has emerged as a powerful 3D representation that harnesses the advantages of both explicit (mesh) and implicit (NeRF) 3D representations. In this paper, we seek to leverage Gaussian splatting to generate realistic animatable avatars from textual descriptions, addressing the limitations (e.g., flexibility and efficiency) imposed by mesh or NeRF-based representations. However, a naive application of Gaussian splatting cannot generate high-quality animatable avatars and suffers from learning instability; it also cannot capture fine avatar geometries and often leads to degenerate body parts. To tackle these problems, we first propose a primitive-based 3D Gaussian representation where Gaussians are defined inside pose-driven primitives to facilitate animation. Second, to stabilize and amortize the learning of millions of Gaussians, we propose to use neural implicit fields to predict the Gaussian attributes (e.g., colors). Finally, to capture fine avatar geometries and extract detailed meshes, we propose a novel SDF-based implicit mesh learning approach for 3D Gaussians that regularizes the underlying geometries and extracts highly detailed textured meshes. Our proposed method, GAvatar, enables the large-scale generation of diverse animatable avatars using only text prompts. GAvatar significantly surpasses existing methods in terms of both appearance and geometry quality, and achieves extremely fast rendering (100 fps) at 1K resolution.
Animation Gallery
| Jack Sparrow | Jack Sparrow | Jack Sparrow | Jack Sparrow |
| Goku | A millitary personnel | A person in a diving suit | Jeff Bezos |
| Kylian Mbappe | Gandalf | A man wearing a hoodie | Serena Williams |
| Optimus prime transformer | Optimus prime transformer | Optimus prime transformer | Optimus prime transformer |
| Shrek | Shrek | Shrek | Shrek |
| Harry Potter | Harry Potter | Harry Potter | Harry Potter |
| Usain Bolt | Usain Bolt | Usain Bolt | Usain Bolt |
Avatar Gallery
| A Viking | Luffy | American Soldier | A man wearing a hoodie |
| A black female surgeon | A clown | A person in a diving suit | Astro boy |
| A bedouin dressed in white | An old man in beige suit | A policewoman | A farmer |
| Ludwig Van Beethoven | Medieval European King | Morty Smith | Serena Williams |
| Kratos | Meghan Markle in a sophisticated outfit | Mobile suit Gundam | Sun Wukong |
Textured Mesh Extraction
| Goku | An old man in beige suit | A professional boxer | A person in a diving suit |
Ablation Study
| w/o SDF field. | w/ SDF field. | w/o SDF field. | w/ SDF field. |
|
w/o Implicit Gaussian Attribute Fields |
w/ Implicit Gaussian Attribute Fields |
w/o Implicit Gaussian Attribute Fields |
w/ Implicit Gaussian Attribute Fields |
Video
Method Overview
Citation
@article{yuan2023gavatar,
title={GAvatar: Animatable 3D Gaussian Avatars with Implicit Mesh Learning},
author={Yuan, Ye and Li, Xueting and Huang, Yangyi and De Mello, Shalini and Nagano, Koki and Kautz, Jan and Iqbal, Umar},
journal={arXiv preprint arXiv:2312.11461},
year={2023}
}
|
Template adapted from GLAMR. |