Research map / Video / Video-language understanding
Scanning Only Once: An End-to-end Framework for Fast Temporal Grounding in Long Videos
Yulin Pan, Xiangteng He, Biao Gong, Yiliang Lv, Yujun Shen, Yuxin Peng, Deli Zhao
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 re-ranking module with frame knowledge to pre-ranking (ablation) | SOONet · MAD | R@1 (IoU=0.3): 7.07 -> 7.65 | Improved |
| add boundary regression module (ablation) | SOONet · MAD | R@1 (IoU=0.3): 7.07 -> 8.52 | Improved |
| combine pre-ranking, re-ranking and boundary regression (ablation) | SOONet · MAD | R@1 (IoU=0.3): 7.07 -> 8.83 | Improved |
| use dual-form approximate rank loss instead of single anchor-rank ApproxNDCG (ablation) | SOONet · MAD | R@1 (IoU=0.3): 8.15 -> 8.83 | Improved |
| use ApproxNDCG ranking loss instead of NCE loss (ablation) | SOONet · MAD | R@1 (IoU=0.3): 4.09 -> 8.15 | Improved |
| use binary cross entropy loss with IoU labels (ablation) | SOONet · MAD | R@1 (IoU=0.3): 8.83 -> 0.01 | Hurt |
| increase base anchor length C0 (ablation) | SOONet · MAD | R@1 | Hurt |
| raise temperature alpha above 0.01 range (ablation) | SOONet · MAD | R@1 | Hurt |
| use end-to-end one-time scanning instead of sliding window | temporal grounding pipeline · MAD | total execution time: 7387.8s -> 505.0s (14.6x vs CLIP) | Improved |
| use SOONet on short-video Ego4d-NLQ vs CONE | temporal grounding · Ego4d-NLQ | R@1 (IoU=0.3): 10.40 -> 8.00 | Hurt |
Closest papers by meaning
- Towards One-to-Many Temporal Grounding (ICML 2026 · similarity 0.91)
- DeCafNet: Delegate and Conquer for Efficient Temporal Grounding in Long Videos (CVPR 2025 · similarity 0.90)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.90)
- SnAG: Scalable and Accurate Video Grounding (CVPR 2024 · similarity 0.89)
- VideoGrounding-DINO: Towards Open-Vocabulary Spatio-Temporal Video Grounding (CVPR 2024 · similarity 0.89)
- Temporal Sentence Grounding with Relevance Feedback in Videos (NeurIPS 2024 · similarity 0.89)
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.
