Research map / Video / Video segmentation and tracking
Visual Alignment Constraint for Continuous Sign Language Recognition
Yuecong Min, Aiming Hao, Xiujuan Chai, 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 visual enhancement (VE) loss (ablation) | CTC baseline CSLR · PHOENIX14 Dev | WER (%): 25.4 -> 23.3 | Improved |
| add visual alignment (VA) distillation loss (ablation) | CTC baseline CSLR · PHOENIX14 Dev | WER (%): 25.4 -> 24.5 | Improved |
| add both VE and VA losses (VAC) (ablation) | CTC baseline CSLR · PHOENIX14 Dev | WER (%): 25.4 -> 21.2 | Improved |
| add both VE and VA losses (VAC) (ablation) | CTC baseline CSLR · PHOENIX14 Test | WER (%): 26.6 -> 22.3 | Improved |
| add VA loss on top of VE loss (ablation) | VE-only model · PHOENIX14 Dev | WER (%): 23.3 -> 21.2 | Improved |
| add BN after each temporal convolution (iteration 1) (ablation) | CSLR training · PHOENIX14 Dev | WER (%): 32.7 -> 27.2 | Improved |
| add BN after each temporal convolution (non-iterative) (ablation) | CSLR training · PHOENIX14 Dev | WER (%): 30.4 -> 25.4 | Improved |
| add iterative training (3 iterations) to BN baseline (ablation) | CSLR training · PHOENIX14 Dev | WER (%): 25.4 -> 24.7 | Improved |
| lower feature-extractor learning rate to 0.1 of alignment model (ablation) | CSLR training · PHOENIX14 Dev | WER (%): 25.4 -> 25.0 | Improved |
| raise feature-extractor learning rate to 10x alignment model (ablation) | CSLR training · PHOENIX14 Dev | WER (%): 25.4 -> 34.8 | Hurt |
| use VA temperature 8 instead of 1 (ablation) | VAC · PHOENIX14 Dev | WER (%): 22.1 -> 21.2 | Improved |
| raise VA loss weight from 25 to 35 (ablation) | VAC · PHOENIX14 Dev | WER (%): 21.2 -> 22.0 | Hurt |
Closest papers by meaning
- Iterative Alignment Network for Continuous Sign Language Recognition (CVPR 2019 · similarity 0.89)
- C2ST: Cross-Modal Contextualized Sequence Transduction for Continuous Sign Language Recognition (ICCV 2023 · similarity 0.88)
- Fully Convolutional Networks for Continuous Sign Language Recognition (ECCV 2020 · similarity 0.88)
- Cross-Sentence Gloss Consistency for Continuous Sign Language Recognition (AAAI 2024 · similarity 0.87)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.85)
- 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.
