Research map / 3D vision / Gaussian splatting
GAvatar: Animatable 3D Gaussian Avatars with Implicit Mesh Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Generates animatable 3D human avatars from text by combining primitive-attached 3D Gaussians, neural implicit attribute fields, and SDF-based mesh regularization trained with SDS.
Problem
Text-to-3D avatar generation with mesh or NeRF representations is limited in geometric detail, rendering efficiency, and flexibility; naive 3D Gaussian splatting is not animatable, unstable to optimize with SDS, and lacks surface geometry, leading to degenerate body parts.
Approach
Attach Gaussians to SMPL-X-driven primitives so they deform with pose, predict Gaussian attributes from canonical-pose neural fields to stabilize SDS optimization, and tie Gaussian opacity to an SDF (NeuS-style kernel) so DMTet can extract and regularize textured meshes.
Key results
- Renders 2.5M-Gaussian avatars at 1024x1024 at 100 fps.
- User study (1512 votes, 42 participants): preferred over DreamGaussian 100% on geometry, AvatarCraft 96.43%, AvatarCLIP 98.81%, and over contemporary TADA 61.90% and DreamHumans 73.81% on geometry quality.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use implicit Gaussian attribute fields instead of directly optimizing Gaussian attributes (ablation) | GAvatar · text-prompted avatar generation | avatar quality and optimization stability | Helped |
| add SDF-based mesh learning to regularize Gaussian-avatar geometry (ablation) | GAvatar · text-prompted avatar generation | geometry quality | Helped |
| use GAvatar rather than AvatarCLIP | text-prompted avatar generation · A/B user study | geometry quality preference: 98.81 | Helped |
| use GAvatar rather than AvatarCLIP | text-prompted avatar generation · A/B user study | appearance quality preference: 97.62 | Helped |
| use GAvatar rather than AvatarCLIP | text-prompted avatar generation · A/B user study | consistency with prompt preference: 97.62 | Helped |
| use GAvatar rather than DreamGaussian | text-prompted avatar generation · A/B user study | geometry quality preference: 100.0 | Helped |
| use GAvatar rather than DreamGaussian | text-prompted avatar generation · A/B user study | appearance quality preference: 98.81 | Helped |
| use GAvatar rather than DreamGaussian | text-prompted avatar generation · A/B user study | consistency with prompt preference: 98.81 | Helped |
Limitations
- Color oversaturation inherited from SDS-based optimization
- Geometry-appearance misalignment, with some geometric detail baked into Gaussian colors
- Loose clothing animation with correct temporal deformation remains challenging
Experimental setup
Datasets: animation database (unnamed)
Benchmarks: 24 text prompts commonly used in baselines (A/B user study)
Metrics: user preference percentage (geometry quality, appearance qua rendering fps
Baselines: DreamGaussian AvatarCLIP AvatarCraft Fantasia3D DreamHumans TADA
Builds on: 3D Gaussian Splatting DreamFusion (SDS) NeuS DMTet SMPL-X
Closest papers by meaning
- 3DGS-Avatar: Animatable Avatars via Deformable 3D Gaussian Splatting (CVPR 2024 · similarity 0.93)
- GASP: Gaussian Avatars with Synthetic Priors (CVPR 2025 · similarity 0.91)
- Human Gaussian Splatting: Real-time Rendering of Animatable Avatars (CVPR 2024 · similarity 0.90)
- GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024 · similarity 0.90)
- GaussianAvatar: Towards Realistic Human Avatar Modeling from a Single Video via Animatable 3D Gaussians (CVPR 2024 · similarity 0.90)
- Animatable Gaussians: Learning Pose-dependent Gaussian Maps for High-fidelity Human Avatar Modeling (CVPR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.