Research map / Video / Video segmentation and tracking
TSPNet: Hierarchical Feature Learning via Temporal Semantic Pyramid for Sign Language Translation
DONGXU LI, Chenchen Xu, Xin Yu, Kaihao Zhang, Benjamin Swift, Hanna Suominen, Hongdong 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 |
|---|---|---|---|
| Use multi-scale segment attention instead of the Conv2d-RNN baseline | Sign language translation on RWTH-PHOENIX-Weather 2014T · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 9.58 -> 13.41 | Improved |
| Jointly model local and non-local video context rather than applying attentions sequentially (ablation) | TSPNet · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 12.97 -> 13.41 | Improved |
| Add multiple segment scales {8, 12, 16} rather than using one scale (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 11.61 -> 13.41 | Improved |
| Include 16-frame segments in the multi-scale representation (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 10.40 -> 13.41 | Improved |
| Add 24-frame segments to the multi-scale representation (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4 | Hurt |
| Replace hierarchical segment aggregation with position-wise pooling (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 13.41 -> 11.13 | Hurt |
| Replace hierarchical segment aggregation with position-wise fully connected aggregation (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 13.41 -> 11.80 | Hurt |
| Allow unrestricted attention across segments instead of restricting attention to semantically relevant neighborhoods (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 13.41 -> 9.35 | Hurt |
| Fine-tune I3D on WSLR datasets (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: 11.23 -> 13.41 | Improved |
| Share positional-embedding weights across scales instead of learning separate layers (ablation) | TSPNet-Joint · RWTH-PHOENIX-Weather 2014T test | BLEU-4: +0.08 BLEU-4 | Improved |
Closest papers by meaning
- Video-Based Sign Language Recognition Without Temporal Segmentation (AAAI 2018 · similarity 0.90)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.90)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.89)
- LLMs are Good Sign Language Translators (CVPR 2024 · similarity 0.89)
- SignGraph: A Sign Sequence is Worth Graphs of Nodes (CVPR 2024 · similarity 0.88)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 2025 · similarity 0.88)
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.
