Research map / Video / Video segmentation and tracking
Self-Mutual Distillation Learning for Continuous Sign Language Recognition
Aiming Hao, Yuecong Min, Xilin Chen
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 weight normalization of the classifier (ablation) | CTC baseline · PHOENIX14 test | WER (%): 25.4 -> 24.7 | Improved |
| add weight normalization of the classifier (ablation) | Baseline+WS · PHOENIX14 test | WER (%): 21.9 -> 21.4 | Improved |
| add weight sharing between visual and contextual classifiers (ablation) | CTC baseline · PHOENIX14 test | WER (%): 24.7 -> 21.4 | Improved |
| add weight sharing between visual and contextual classifiers (ablation) | CTC baseline · PHOENIX14 dev | WER (%): 24.1 -> 21.2 | Improved |
| update shared classifier with both LVFs and GCFs instead of only one (ablation) | SMKD weight sharing · PHOENIX14 test | WER (%): 22.5 -> 21.4 | Improved |
| add GSBA gloss segmentation to the visual module (ablation) | CTC baseline · PHOENIX14 test | WER (%): 24.7 -> 22.0 | Improved |
| use soft-target KD from contextual module instead of WS (ablation) | CTC baseline · PHOENIX14 test | WER (%): 21.4 -> 23.6 | Hurt |
| share classifier during synchronous training rather than independent classifiers (ablation) | Baseline · PHOENIX14 test | WER (%): 22.4 -> 21.4 | Improved |
| add gloss segmentation stage after sync training (ablation) | Baseline+sync+dec train · PHOENIX14 test | WER (%): 21.3 -> 21.0 | Improved |
| add decouple training stage at the end (ablation) | Baseline+sync+gloss segment · PHOENIX14 test | WER (%): 21.3 -> 21.0 | Improved |
| add decouple training stage after sync only (ablation) | Baseline+sync · PHOENIX14 test | WER (%): 21.4 -> 21.3 | No clear effect |
| vary alpha balancing visual and contextual CTC losses (ablation) | SMKD weight sharing · PHOENIX14 | WER (%): optimal alpha 0.5 | Mixed |
Closest papers by meaning
- Distilling Cross-Temporal Contexts for Continuous Sign Language Recognition (CVPR 2023 · similarity 0.91)
- Recurrent Convolutional Neural Networks for Continuous Sign Language Recognition by Staged Optimization (CVPR 2017 · similarity 0.91)
- C2ST: Cross-Modal Contextualized Sequence Transduction for Continuous Sign Language Recognition (ICCV 2023 · similarity 0.90)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.89)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.88)
- Cross-Sentence Gloss Consistency for Continuous Sign Language Recognition (AAAI 2024 · 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.
