Research map / 3D vision / Face and shape
Generalizable One-shot 3D Neural Head Avatar
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the appearance branch to transfer source-image details into the 3D representation (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA 3D portrait reconstruction | FID: 47.84 -> 10.47 | Helped |
| Add the appearance branch to preserve source identity during reenactment (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | CSIM: 0.370 -> 0.599 | Helped |
| Add a neutral-expression constraint to the canonical reconstruction (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | AED: 0.315 -> 0.276 | Helped |
| Use a frontal 3DMM-rendered target expression instead of linearly combined expression bases (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | FID: 26.18 -> 17.36 | Helped |
| Use a frontal 3DMM-rendered expression rather than StyleGAN2 modulation from expression coefficients (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | CSIM: 0.558 -> 0.599 | Helped |
| Use a frontal 3DMM-rendered expression rather than directly encoding expression coefficients (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | CSIM: 0.538 -> 0.599 | Helped |
| Fine-tune only the super-resolution module in stage II rather than fine-tuning end-to-end (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA cross-identity reenactment | FID: 21.51 -> 17.36 | Helped |
| Fine-tune the full model end-to-end in stage II (ablation) | Generalizable One-shot 3D Neural Head Avatar · CelebA 3D portrait reconstruction | PSNR: 24.77 -> 25.87 | Helped |
| Use the proposed three-branch representation and rendering rather than ROME | Generalizable One-shot 3D Neural Head Avatar · CelebA 3D portrait reconstruction | LPIPS: 0.085 -> 0.040 | Helped |
| Use the proposed method rather than ROME | Generalizable One-shot 3D Neural Head Avatar · HDTF same-identity reenactment | PSNR: 20.75 -> 22.15 | Helped |
| Use the proposed method rather than ROME | Generalizable One-shot 3D Neural Head Avatar · HDTF cross-identity reenactment | CSIM: 0.629 -> 0.643 | Helped |
| Use the proposed method rather than Next3D-PTI | Generalizable One-shot 3D Neural Head Avatar · HDTF to CelebA cross-identity reenactment | CSIM: 0.483 -> 0.551 | Helped |
Closest papers by meaning
- Generalizable and Animatable Gaussian Head Avatar (NeurIPS 2024 · similarity 0.91)
- Synthetic Prior for Few-Shot Drivable Head Avatar Inversion (CVPR 2025 · similarity 0.90)
- GPAvatar: Generalizable and Precise Head Avatar from Image(s) (ICLR 2024 · similarity 0.90)
- Efficient 3D Implicit Head Avatar with Mesh-anchored Hash Table Blendshapes (CVPR 2024 · similarity 0.90)
- Arc2Avatar: Generating Expressive 3D Avatars from a Single Image via ID Guidance (CVPR 2025 · similarity 0.90)
- High-Fidelity 3D Head Avatars Reconstruction through Spatially-Varying Expression Conditioned Neural Radiance Field (AAAI 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.