Research map / 3D vision / Face and shape
GeoAvatar: Adaptive Geometrical Gaussian Splatting for 3D Head Avatar
SeungJun Moon, Hah Min Lew, Seungeun Lee, Ji-Su Kang, Gyeong-Moon Park
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Adaptive Pre-allocation Stage (rigid/flexible split) to the baseline (ablation) | GaussianAvatars_0 baseline · SplattingAvatar+DynamicFace ablation | LPIPS (1e-1): 0.653 -> 0.572 | Improved |
| add extended FLAME mouth structure (teeth, palate, floor) (ablation) | GaussianAvatars_0 + APS · SplattingAvatar+DynamicFace ablation | PSNR: 30.717 -> 32.273 | Improved |
| add part-wise mouth deformation network (ablation) | GaussianAvatars_0 + APS + FLAME mouth · SplattingAvatar+DynamicFace ablation | LPIPS (1e-1): 0.548 -> 0.519 | Improved |
| add L_angle regularization on the polar angle of Gaussians (ablation) | GeoAvatar (without L_angle) · SplattingAvatar+DynamicFace ablation | PSNR: 32.751 -> 32.697 | Mixed |
| add APS on top of FLAME mouth (versus all-flexible initialization) (ablation) | GeoAvatar · SplattingAvatar+DynamicFace ablation | PSNR: 31.318 -> 32.697 | Improved |
| use APS instead of treating all regions as rigid (ablation) | GeoAvatar with FLAME mouth · SplattingAvatar+DynamicFace ablation | LPIPS (1e-1): 0.840 -> 0.513 | Improved |
| APS or FLAME mouth alone versus together (ablation) | GeoAvatar · SplattingAvatar+DynamicFace ablation | PSNR: 30.356 -> 32.697 | Improved |
| train GeoAvatar on a single view versus GaussianAvatars on 16 views | GeoAvatar · NeRSemble (12 actors) | LPIPS (1e-1): 1.073 -> 0.969 | Improved |
| train GeoAvatar on single view versus GaussianAvatars on 16 views, PSNR | GeoAvatar · NeRSemble (12 actors) | PSNR: 27.829 -> 27.782 | No clear effect |
| use GeoAvatar instead of best baseline on DynamicFace | GeoAvatar · DynamicFace | PSNR: 31.026 -> 32.760 | Improved |
| GeoAvatar costs more training time and slower inference than FlashAvatar | GeoAvatar · Efficiency (RTX 3090) | Inference speed (FPS): 291.20 -> 71.52 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FlashAvatar — FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding (CVPR 2024) · better on 6 of 6 shared results
- GaussianAvatars_0 — GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025 · similarity 0.91)
- GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024 · similarity 0.90)
- SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting (CVPR 2024 · similarity 0.89)
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.88)
- HRAvatar: High-Quality and Relightable Gaussian Head Avatar (CVPR 2025 · similarity 0.88)
- GeoAvatar: Geometrically-Consistent Multi-Person Avatar Reconstruction from Sparse Multi-View Videos (CVPR 2025 · similarity 0.88)
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.
