Research map / 3D vision / Human pose estimation
MySign: A High-Fidelity Motion-Capture Dataset for 3D Sign Generation in Bahasa Isyarat Malaysia
Jiayu Shen, Kalin Stefanov, Lay-Ki Soon, Vee Yee CHONG, Alpha A Gopalai, KokSheik Wong
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces MySign, a motion-capture dataset of 5,000 3D isolated sign samples covering 1,000 glosses in Bahasa Isyarat Malaysia, performed by five Deaf native signers.
Problem
Fine-grained 3D sign generation is limited by the scarcity of high-fidelity 3D sign language data, especially for under-resourced languages such as Bahasa Isyarat Malaysia (BIM).
Approach
A motion-capture dataset of isolated BIM signs with 5,000 3D samples over 1,000 glosses, recorded from five Deaf native signers, positioned as higher fidelity than prior sign datasets.
Key results
- MySign comprises 5,000 3D motion-capture samples covering 1,000 lexical items (glosses) in Bahasa Isyarat Malaysia.
- Data was produced by five Deaf native signers.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on MoCap-grounded SMPL-X instead of RGB-fitted pseudo-labels (ablation) | wSignGen · MySign | FID to MoCap-real: 51.400 -> 42.800 | Improved |
| train on MoCap-grounded instead of RGB-fitted supervision (ablation) | wSignGen · MySign | Acc: 0.704 -> 0.747 | Improved |
| use sign-specific generator (wSignGen) instead of adapted general motion model (MotionDiffuse) | MotionDiffuse · MySign | FID: 83.564 -> 41.465 | Improved |
| pretrain on MySign before fine-tuning on ASL3DWord (ablation) | SignAvatar · ASL3DWord test | FID: 71.809 -> 63.831 | Improved |
| pretrain on ASL3DWord before fine-tuning on MySign (ablation) | SignAvatar · MySign test | FID: 54.166 -> 65.037 | Hurt |
| pretrain on ASL3DWord before fine-tuning on MySign (ablation) | SignAvatar · MySign test | Acc: 0.727 -> 0.581 | Hurt |
| apply hand ROM clamping to retargeted motion (ablation) | ST-GCN evaluator · MySign test | Top-1 (%): 85.0 -> 85.7 | No clear effect |
| stronger SLR model (SignBERT+ vs ST-GCN) | Recognition baselines · MySign | Top-1 (%): 85.70 -> 92.90 | Improved |
Closest papers by meaning
- SignAvatars: A Large-scale 3D Sign Language Holistic Motion Dataset and Benchmark (ECCV 2024 · similarity 0.92)
- FSboard: Over 3 Million Characters of ASL Fingerspelling Collected via Smartphones (CVPR 2025 · similarity 0.87)
- Neural Sign Actors: A Diffusion Model for 3D Sign Language Production from Text (CVPR 2024 · similarity 0.87)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.87)
- MM-WLAuslan: Multi-View Multi-Modal Word-Level Australian Sign Language Recognition Dataset (NeurIPS 2024 · similarity 0.87)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 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.
