Research map / Graph learning / Knowledge and temporal graphs
HyperSign: Hierarchical Hypergraph-based Co-occurrence Modeling for Sign Language Recognition and Translation
Qianren Guo, Yuehang Wang, Yongji Zhang, Qi Chu, Sen Liu, Yu Jiang
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 the CGP module (physical graph plus geometric and semantic hypergraphs) (ablation) | skeleton graph-conv baseline · PHOENIX-2014T SLR | WER (%): 27.3 / 28.1 -> 20.8 / 21.1 | Improved |
| add the MPHF module for cross-part hypergraph fusion (ablation) | Baseline+CGP · PHOENIX-2014T SLR | WER (%): 20.8 / 21.1 -> 18.9 / 19.7 | Improved |
| add UACD uncertainty-aware collaborative distillation (ablation) | Baseline+CGP+MPHF · PHOENIX-2014T SLR | WER (%): 18.9 / 19.7 -> 18.6 / 19.2 | Improved |
| increase KNN neighbors K from 8 to 9 (P=8) (ablation) | HyperSign CGP · PHOENIX-2014T SLR | WER (%): 18.6 / 19.2 -> 19.0 / 20.1 | Hurt |
| increase prototypes P from 8 to 9 (K=8) (ablation) | HyperSign CGP · PHOENIX-2014T SLR | WER (%): 18.6 / 19.2 -> 18.7 / 20.0 | Hurt |
| decrease K from 8 to 7 (P=8) (ablation) | HyperSign CGP · PHOENIX-2014T SLR | WER (%): 18.6 / 19.2 -> 18.7 / 19.3 | Hurt |
| decrease P from 8 to 7 (K=8) (ablation) | HyperSign CGP · PHOENIX-2014T SLR | WER (%): 18.6 / 19.2 -> 18.6 / 19.5 | Hurt |
| use HyperSign instead of MSKA | skeleton SLR · CSL-Daily SLR | WER (%): 27.5 / 27.1 -> 26.1 / 25.7 | Improved |
| use HyperSign instead of MSKA | skeleton SLT · PHOENIX-2014T | BLEU4: 29.03 -> 29.35 | Improved |
| use HyperSign instead of MSKA | skeleton SLT · CSL-Daily | BLEU4: 25.52 -> 26.87 | Improved |
| use HyperSign instead of MSKA (efficiency) | skeleton SLR · PHOENIX-2014T | FPS: 1224 -> 1397 | Improved |
| skeleton HyperSign versus RGB CV-SLT | SLT · PHOENIX-2014T TEST | ROUGE: 54.26 vs. 54.54 | No clear effect |
Closest papers by meaning
- HyperSign: Saliency-Aware Spatial Graphs and Temporal Hypergraphs for Continuous Sign Language Recognition (AAAI 2026 · similarity 0.93)
- Geo-Sign: Hyperbolic Contrastive Regularisation for Geometrically Aware Sign Language Translation (NeurIPS 2025 · similarity 0.88)
- SignGraph: A Sign Sequence is Worth Graphs of Nodes (CVPR 2024 · similarity 0.88)
- VSNet: Focusing on the Linguistic Characteristics of Sign Language (CVPR 2025 · similarity 0.87)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.86)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.85)
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.
