Research map / Label-efficient and robust learning / Recognition and long-tail
Stochastic Fine-grained Labeling of Multi-state Sign Glosses for Continuous Sign Language Recognition
Zhe Niu, Brian Mak
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 stochastic frame dropping (ablation) | CSLR baseline · PHOENIX-2014 Test | WER (%): 33.9 -> 30.3 | Improved |
| add stochastic frame dropping (ablation) | CSLR baseline · PHOENIX-2014 Dev | WER (%): 34.1 -> 29.1 | Improved |
| add stochastic gradient stopping (pstop 75%) on top of SFD (ablation) | SFD model · PHOENIX-2014 Dev | WER (%): 29.1 -> 29.8 | Hurt |
| add stochastic gradient stopping on top of SFD (ablation) | SFD model · PHOENIX-2014 Test | WER (%): 30.3 -> 29.4 | Improved |
| add deterministic fine-grained labeling with 2 states (ablation) | SFD+SGS · PHOENIX-2014 Dev | WER (%): 29.8 -> 26.5 | Improved |
| add deterministic fine-grained labeling with 2 states (ablation) | SFD+SGS · PHOENIX-2014 Test | WER (%): 29.4 -> 27.2 | Improved |
| replace deterministic labeling with stochastic fine-grained labeling (Cmax 5) (ablation) | SFD+SGS+DFL · PHOENIX-2014 Dev | WER (%): 26.5 -> 26.2 | Improved |
| replace deterministic labeling with stochastic fine-grained labeling (Cmax 5) (ablation) | SFD+SGS+DFL · PHOENIX-2014 Test | WER (%): 27.2 -> 26.8 | Improved |
| train state-number distribution with REINFORCE instead of uniform (ablation) | SFL, Cmax=5 · PHOENIX-2014 Dev | WER (%): 30.2 -> 26.2 | Improved |
| train state-number distribution with REINFORCE instead of uniform (ablation) | SFL, Cmax=5 · PHOENIX-2014 Test | WER (%): 30.3 -> 26.8 | Improved |
| add the dataset language model during decoding | SFD+SGS+SFL · PHOENIX-2014 Dev | WER (%): 26.2 -> 24.9 | Improved |
| add the dataset language model during decoding | SFD+SGS+SFL · PHOENIX-2014 Test | WER (%): 26.8 -> 25.3 | Improved |
Closest papers by meaning
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.87)
- Cross-Sentence Gloss Consistency for Continuous Sign Language Recognition (AAAI 2024 · similarity 0.86)
- Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · similarity 0.86)
- Improving Gloss-free Sign Language Translation by Reducing Representation Density (NeurIPS 2024 · similarity 0.86)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.85)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · 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.
