Research map / 3D vision / Face and shape
Feed-forward Gaussian Registration for Head Avatar Creation and Editing
Malte Prinzler, Paulo Gotardo, Siyu Tang, Timo Bolkart
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 |
|---|---|---|---|
| use registration-guided attention instead of dense attention (ablation) | MATCH · Ava-256 (300k iterations) | LPIPS: 0.221 -> 0.187 | Improved |
| add pretrained Sapiens features (ablation) | MATCH · Ava-256 (300k iterations) | LPIPS: 0.202 -> 0.187 | Improved |
| add skip connections for color and location (ablation) | MATCH · Ava-256 (300k iterations) | LPIPS: 0.192 -> 0.187 | Improved |
| use adapted TEMPEH global stage instead of original TEMPEH for coarse mesh (ablation) | MATCH · Ava-256 (300k iterations) | LPIPS: 0.190 -> 0.187 | Improved |
| increase UV texture resolution from 256 to 1024 (ablation) | MATCH · Ava-256 (300k iterations) | LPIPS: 0.194 (256) -> 0.190 (512) -> 0.187 | Improved |
| smaller k of image tokens to attend to in registration-guided attention (ablation) | MATCH · Ava-256 | reconstruction quality | Improved |
| joint training on Ava-256 and NeRSemble without NeRSemble geometry supervision (ablation) | MATCH · NeRSemble | LPIPS: 0.168 (NeRSemble only) -> 0.152 | Improved |
| train on Ava-256 only and test on unseen NeRSemble (ablation) | MATCH · NeRSemble | LPIPS: FaceLift 0.200 vs Ava-256 only 0.182 | Improved |
| build GEM avatars from MATCH-predicted Gaussians | GEM · Ava-256 subject-specific avatars | reconstruction time: 45.3h -> 4.6h | Improved |
| build GEM avatars from MATCH-predicted Gaussians | GEM · Ava-256 cross-reenactment | EmoL1: 9.866 -> 9.837 | No clear effect |
| use MATCH geometry instead of TEMPEH in mouth and eye regions | MATCH · Ava-256 geometry (eyes) | P2S (mm): Mouth P2S 0.69 -> 0.72; Eyes P2S 0.62 -> 0.61 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- FaceLift — FaceLift: Learning Generalizable Single Image 3D Face Reconstruction from Synthetic Heads (ICCV 2025) · better on 3 of 3 shared results
- GPAvatar — GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025) · better on 1 of 1 shared result
- RGBAvatar — RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head Avatars (CVPR 2025) · better on 1 of 1 shared result
- Avat3r — Avat3r: Large Animatable Gaussian Reconstruction Model for High-fidelity 3D Head Avatars (ICCV 2025) · better on 1 of 1 shared result
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
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.91)
- MeGA: Hybrid Mesh-Gaussian Head Avatar for High-Fidelity Rendering and Head Editing (CVPR 2025 · similarity 0.90)
- GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view Diffusion (CVPR 2025 · similarity 0.89)
- GASP: Gaussian Avatars with Synthetic Priors (CVPR 2025 · similarity 0.89)
- Arc2Avatar: Generating Expressive 3D Avatars from a Single Image via ID Guidance (CVPR 2025 · similarity 0.89)
- Gaussian Eigen Models for Human Heads (CVPR 2025 · similarity 0.89)
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.
