Research map / 3D vision / Face and shape
PHOSA: Photorealistic 3D Sign Avatar Modeling and Benchmark
Haodong Wang, Hezhen Hu, Wengang Zhou, Houqiang Li
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 hybrid SMPL-X fitting (HaMeR hands, INFERNO face) instead of DWPose-only fitting (ablation) | avatar modeling · MVSign (Hand region) | PSNR: 16.82 -> 18.85 | Improved |
| add hybrid SMPL-X fitting (ablation) | avatar modeling · MVSign (Full body) | PSNR: 23.58 -> 26.90 | Improved |
| increase training views from 1 to 16 (ablation) | PHOSA · MVSign (Hand region) | PSNR: 18.49 -> 18.85 | Improved |
| train on basic hand shapes only instead of both subsets (ablation) | PHOSA · MVSign (Hand region) | PSNR: 18.85 -> 17.47 | Hurt |
| train on sign sentences only instead of both subsets (ablation) | PHOSA · MVSign (Hand region) | PSNR: 18.85 -> 17.62 | Hurt |
| use motion-aware sampling instead of random sampling (ablation) | PHOSA · MVSign (Hand region) | PSNR: 17.40 -> 18.03 | Improved |
| use motion-aware sampling instead of sequential sampling (ablation) | PHOSA · MVSign (Full body) | PSNR: 26.55 -> 27.67 | Improved |
| decouple body, head and hand Gaussian maps (ablation) | PHOSA · MVSign (Hand region) | PSNR: 16.65 -> 17.53 | Improved |
| add partial kinematic decoupling for hands on top of decoupling (ablation) | PHOSA · MVSign (Hand region) | PSNR: 17.53 -> 18.74 | Improved |
| PHOSA vs best baseline on in-the-wild monocular video | PHOSA · In-the-wild monocular web video (Full body) | PSNR: 19.61 -> 20.39 | Improved |
| PHOSA vs best baseline on MVSign | PHOSA · MVSign multi-view (Full body) | PSNR: 25.91 -> 27.03 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GaussianAvatar — GaussianAvatar: Towards Realistic Human Avatar Modeling from a Single Video via Animatable 3D Gaussians (CVPR 2024) · 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
- Reconstructing Signing Avatars From Video Using Linguistic Priors (CVPR 2023 · similarity 0.89)
- MySign: A High-Fidelity Motion-Capture Dataset for 3D Sign Generation in Bahasa Isyarat Malaysia (NeurIPS 2026 · similarity 0.88)
- SignAvatars: A Large-scale 3D Sign Language Holistic Motion Dataset and Benchmark (ECCV 2024 · similarity 0.88)
- Neural Sign Actors: A Diffusion Model for 3D Sign Language Production from Text (CVPR 2024 · similarity 0.86)
- From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations (CVPR 2024 · similarity 0.84)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 2025 · similarity 0.84)
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.
