Research map / Label-efficient and robust learning / Self-supervised and contrastive
SignBERT: Pre-Training of Hand-Model-Aware Representation for Sign Language Recognition
Hezhen Hu, Weichao Zhao, Wengang Zhou, Yuechen Wang, 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 self-supervised pre-training (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 63.01 -> 76.09 | Improved |
| add self-supervised pre-training (ablation) | SignBERT (H) · MSASL1000 | Top-1 per-instance accuracy: 41.85 -> 49.54 | Improved |
| add masked joint modeling (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 63.01 -> 72.66 | Improved |
| add masked frame modeling (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 63.01 -> 74.77 | Improved |
| add masked frame modeling on top of joint masking (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 72.66 -> 76.09 | Improved |
| add masked joint modeling on top of frame masking (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 74.77 -> 76.09 | Improved |
| use model-aware MANO decoder instead of 1-layer fc decoder (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 73.05 -> 76.09 | Improved |
| use model-aware MANO decoder instead of 2-layer fc decoder (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 74.24 -> 76.09 | Improved |
| use model-aware MANO decoder instead of 1-layer fc decoder (ablation) | SignBERT (H) · MSASL1000 | Top-1 per-instance accuracy: 47.94 -> 49.54 | Improved |
| use 3 Transformer layers instead of 2 (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 74.11 -> 76.09 | Improved |
| use 3 Transformer layers instead of 4 (ablation) | SignBERT (H) · MSASL1000 | Top-1 per-instance accuracy: 47.36 -> 49.54 | Improved |
| use 3 Transformer layers instead of 5 (ablation) | SignBERT (H) · MSASL100 | Top-1 per-instance accuracy: 74.90 -> 76.09 | Improved |
Closest papers by meaning
- SignBERT+: Hand-Model-Aware Self-Supervised Pre-Training for Sign Language Understanding (TPAMI 2023 · similarity 0.98)
- Hand-Model-Aware Sign Language Recognition (AAAI 2021 · similarity 0.94)
- SignRep: Enhancing Self-Supervised Sign Representations (ICCV 2025 · similarity 0.90)
- Uni-Sign: Toward Unified Sign Language Understanding at Scale (ICLR 2025 · similarity 0.87)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.87)
- Improving Gloss-free Sign Language Translation by Reducing Representation Density (NeurIPS 2024 · similarity 0.86)
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.
