Research map / Video / Video generation
SignRefine: Adapting Foundational Video Models for Sign Language Generation
Anton Pelykh, Edward Fish, Ozge Mercanoglu Sincan, Richard Bowden
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 spatially grounded local adapters for hands and face | Wan2.1-1.3B-Fun-Control · NVSign test (Right Hand) | MPJPE: 9.565 -> 6.602 | Improved |
| add local adapters | Wan2.1-1.3B-Fun-Control · BSL Corpus (Face) | LPIPS: 0.221 -> 0.222 | No clear effect |
| add local adapters | Wan2.1-1.3B-Fun-Control · User study (overall) | mean rank: 1.89 -> 1.22 | Improved |
| add local adapters | Wan2.1-1.3B-Fun-Control · Comprehension study | Keyword Recall: 0.30 -> 0.54 | Improved |
| add attention masking with sink token (ablation) | SignRefine · NVSign test (Face) | MPJPE: 1.949 -> 1.770 | Improved |
| add attention masking with sink token (ablation) | SignRefine · NVSign test (Left Hand) | MPJPE: 7.309 -> 6.638 | Improved |
| add CoordConv regional coordinates (ablation) | SignRefine · NVSign test (Face) | MPJPE: 1.847 -> 1.770 | Improved |
| use dense conditions (face normals, MANO) instead of sparse keypoints (ablation) | SignRefine adapters · NVSign test (Face) | MPJPE: 1.770 -> 1.692 | Improved |
| use dense conditions instead of sparse keypoints (ablation) | SignRefine adapters · NVSign test (Right Hand) | MPJPE: 6.602 -> 6.713 | No clear effect |
Closest papers by meaning
- SignGen: End-to-End Sign Language Video Generation with Latent Diffusion (ECCV 2024 · similarity 0.93)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 2025 · similarity 0.91)
- Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · similarity 0.90)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.89)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.88)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.88)
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.
