Research map / Video / Video-language understanding
Find and Focus: Retrieve and Localize Video Events with Natural Language Queries
Dian Shao, Yu Xiong, Yue Zhao, Qingqiu Huang, Yu Qiao, Dahua Lin
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 Focus-stage part-level refinement in top 100 to the Find stage | VSE · ActivityNet Captions val_1 (video retrieval) | Recall@1: 11.69 -> 14.05 | Improved |
| add Focus-stage refinement in top 20 | VSE · Modified LSMDC (video retrieval) | Recall@1: 2.66 -> 3.89 | Improved |
| add Focus-stage refinement | VSE · Modified LSMDC (video retrieval) | Recall@5: 10.63 -> 13.70 | Improved |
| use semantic-relevance proposals (VSS) instead of SSN+shot (ablation) | clip localization · ActivityNet Captions (clip localization) | Recall@tIoU=0.3: 16.71 -> 28.52 | Improved |
| use tf-idf weighting in BoW text features (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 13.46 -> 14.05 | Improved |
| use Fisher vector word aggregation instead of plain word2vec (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 10.92 -> 13.75 | Improved |
| add feature refinement in Focus stage (ablation) | VSS clip localization · ActivityNet Captions (clip localization) | Recall@tIoU=0.5: 12.74 -> 13.46 | Improved |
| add feature refinement in Focus stage (ablation) | VSS clip localization · Modified LSMDC (clip localization) | Recall@tIoU=0.5: 2.48 -> 2.49 | No clear effect |
| increase K from 20 to 100 (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 14.11 -> 14.05 | No clear effect |
| increase K from 100 to 1000 (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 14.05 -> 14.01 | No clear effect |
| raise umax from 1 to 2 in bipartite matching (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 13.87 -> 13.93 | Improved |
| raise umax from 2 to 3 in bipartite matching (ablation) | FIFO · ActivityNet Captions val_1 (video retrieval) | Recall@1: 13.93 -> 13.75 | Hurt |
Closest papers by meaning
- Localizing Events in Videos with Multimodal Queries (CVPR 2025 · similarity 0.89)
- Multilevel Language and Vision Integration for Text-to-Clip Retrieval (AAAI 2019 · similarity 0.88)
- TALL: Temporal Activity Localization via Language Query (ICCV 2017 · similarity 0.88)
- MultiVENT 2.0: A Massive Multilingual Benchmark for Event-Centric Video Retrieval (CVPR 2025 · similarity 0.86)
- Do You Remember? Dense Video Captioning with Cross-Modal Memory Retrieval (CVPR 2024 · similarity 0.86)
- Fewer Steps, Better Performance: Efficient Cross-Modal Clip Trimming for Video Moment Retrieval Using Language (AAAI 2024 · similarity 0.86)
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.
