Research map / Video / Video-language understanding
Multi-Query Video Retrieval
Zeyu Wang, Yu Wu, Karthik Narasimhan, Olga Russakovsky
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 |
|---|---|---|---|
| aggregate similarity instead of ranks in post-hoc multi-query inference | CLIP4Clip single-query model · MSR-VTT | R@1: 56.4 -> 68.4 | Improved |
| aggregate similarity instead of ranks in post-hoc multi-query inference | CLIP4Clip single-query model · MSVD | R@1: 47.3 -> 57.6 | Improved |
| aggregate similarity instead of ranks in post-hoc multi-query inference | CLIP4Clip single-query model · VATEX | R@1: 43.6 -> 47.3 | Improved |
| train with five-query mean feature instead of single-query training with similarity aggregation (ablation) | CLIP4Clip · MSR-VTT | R@1: 68.4 -> 71.3 | Improved |
| train with five-query mean feature instead of single-query training with similarity aggregation (ablation) | CLIP4Clip · VATEX | R@1: 47.3 -> 57.2 | Improved |
| train with five-query mean feature instead of single-query training with similarity aggregation (ablation) | CLIP4Clip · MSVD | R@1: 57.6 -> 59.6 | Improved |
| add contextualized weight generation to mean feature training (ablation) | CLIP4Clip MF · MSR-VTT | R@1: 71.3 -> 73.1 | Improved |
| add contextualized weight generation to mean feature training (ablation) | CLIP4Clip MF · MSVD | R@1: 59.6 -> 61.7 | Improved |
| add contextualized weight generation to mean feature training (ablation) | CLIP4Clip MF · VATEX | R@1: 57.2 -> 59.0 | Improved |
| use contextualized instead of localized weight generation (ablation) | CLIP4Clip WF · MSVD | R@1: 60.2 -> 61.7 | Improved |
| add text-to-text similarity weighting to mean feature on Frozen (ablation) | Frozen MF · MSR-VTT | R@1: 59.6 -> 60.6 | Improved |
| train with five queries instead of single query (single-query test) | CLIP4Clip CG-WF vs SA · MSR-VTT | R@1 with 1 query: 41.5 -> 41.0 | Hurt |
Closest papers by meaning
- MUSE: Mamba Is Efficient Multi-scale Learner for Text-video Retrieval (AAAI 2025 · similarity 0.87)
- MUVR: A Multi-Modal Untrimmed Video Retrieval Benchmark with Multi-Level Visual Correspondence (NeurIPS 2025 · similarity 0.87)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.87)
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.86)
- Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval (CVPR 2025 · similarity 0.86)
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · 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.
