Research map / 3D vision / Face and shape
Realistic One-Shot Mesh-Based Head Avatars
Taras Khakhulin, Vanessa Sklyarova, Victor Lempitsky, Egor Zakharov
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 learned per-vertex displacements to the FLAME mesh (ablation) | ROME · VoxCeleb2 hold-out (ablation) | LPIPS: 0.10 -> 0.08 | Improved |
| add learned per-vertex displacements to the FLAME mesh (ablation) | ROME · VoxCeleb2 hold-out (ablation) | SSIM: 0.81 -> 0.86 | Improved |
| add learned per-vertex displacements to the FLAME mesh (ablation) | ROME · VoxCeleb2 hold-out (ablation) | PSNR: 23.1 -> 25.8 | Improved |
| displace vertices along normals instead of free deformation (ablation) | ROME · VoxCeleb2 (qualitative) | render quality | Improved |
| add silhouette-based occupancy loss (ablation) | ROME · VoxCeleb2 (qualitative) | reconstruction | Improved |
| add Laplacian regularization (ablation) | ROME · VoxCeleb2 (qualitative) | smoothness | Improved |
| add 2D Chamfer loss (ablation) | ROME · VoxCeleb2 (qualitative) | smoothness of visible reconstruction | Improved |
| distill offsets into a PCA linear basis (50 hair, 10 neck) | LinearROME · H3DS one-shot reconstruction | Chamfer distance: 12.6 -> 12.5 | No clear effect |
| use ROME instead of H3D-Net for one-shot reconstruction | ROME · H3DS one-shot reconstruction | Chamfer distance: 15.1 -> 12.6 | Improved |
| use ROME instead of TPSMM in cross-reenactment | ROME · VoxCeleb2-HQ cross-reenactment | FID: 49.27 -> 45.32 | Improved |
| use ROME instead of TPSMM in self-reenactment | ROME · VoxCeleb2-HQ self-reenactment | PSNR: 26.1 -> 26.2 | No clear effect |
Closest papers by meaning
- Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars (ECCV 2020 · similarity 0.92)
- Synthetic Prior for Few-Shot Drivable Head Avatar Inversion (CVPR 2025 · similarity 0.88)
- Generalizable and Animatable Gaussian Head Avatar (NeurIPS 2024 · similarity 0.88)
- HERA: Hybrid Explicit Representation for Ultra-Realistic Head Avatars (CVPR 2025 · similarity 0.88)
- Efficient 3D Implicit Head Avatar with Mesh-anchored Hash Table Blendshapes (CVPR 2024 · similarity 0.87)
- MeGA: Hybrid Mesh-Gaussian Head Avatar for High-Fidelity Rendering and Head Editing (CVPR 2025 · similarity 0.87)
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.
