Research map / 3D vision / Face and shape
FFAvatar: Feed-Forward 4D Head Avatar Reconstruction from Arbitrary Images
Jianjiang Yao, Ke Xian, Renxiang Dai, Robert Qiu
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 motion refinement module (MARM) (ablation) | FFAvatar · VFHQ novel expressions (few-shot, 8 inputs) | PSNR: 23.20 -> 23.35 | Improved |
| add motion refinement module (MARM) (ablation) | FFAvatar · VFHQ novel expressions (few-shot, 8 inputs) | LPIPS: 0.088 -> 0.081 | Improved |
| add motion refinement module (MARM) (ablation) | FFAvatar · VFHQ novel expressions (few-shot, 8 inputs) | FPS: 468 -> 31 | Hurt |
| increase input frames from 1 to 8 (ablation) | FFAvatar · VFHQ novel expressions, input count | PSNR: 21.82 -> 23.35 | Improved |
| increase input frames from 8 to 32 (ablation) | FFAvatar · VFHQ novel expressions, input count | PSNR: 23.35 -> 23.38 | No clear effect |
| use sparse-to-dense instead of sparse-only (S-5K) (ablation) | FFAvatar · Sparse-to-dense ablation (UV 128) | PSNR: 19.69 -> 23.35 | Improved |
| use sparse-to-dense instead of direct dense optimization (D-20K) (ablation) | FFAvatar · Sparse-to-dense ablation (UV 128) | train time per 1000 iters: 120min -> 47min | Improved |
| use sparse-to-dense instead of direct dense optimization (D-20K) (ablation) | FFAvatar · Sparse-to-dense ablation (UV 128) | PSNR: 23.41 -> 23.35 | No clear effect |
| direct dense optimization at UV 256 (D-64K) (ablation) | FFAvatar · Sparse-to-dense ablation (UV 256) | feasibility | Hurt |
| FFAvatar vs GPAvatar few-shot on NeRSemble novel views | FFAvatar · NeRSemble novel views (few-shot) | LPIPS: 0.163 -> 0.098 | Improved |
| FFAvatar one-shot vs GAGAvatar/LAM PSNR | FFAvatar · VFHQ novel expressions (one-shot) | PSNR: 21.82 vs LAM 22.65 | Mixed |
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.
- FastAvatar — FastAvatar: Towards Unified and Fast 3D Avatar Reconstruction with Large Gaussian Reconstruction Transformers (ICLR 2026) · better on 4 of 4 shared results
- GPAvatar — GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025) · better on 4 of 4 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
- FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding (CVPR 2024 · similarity 0.91)
- RGBAvatar: Reduced Gaussian Blendshapes for Online Modeling of Head Avatars (CVPR 2025 · similarity 0.90)
- HRAvatar: High-Quality and Relightable Gaussian Head Avatar (CVPR 2025 · similarity 0.90)
- GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025 · similarity 0.90)
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.90)
- GPAvatar: Generalizable and Precise Head Avatar from Image(s) (ICLR 2024 · 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.
