Research map / Video / Video-language understanding
Text-Adaptive Multiple Visual Prototype Matching for Video-Text Retrieval
Chengzhi Lin, Ancong Wu, Junwei Liang, Jun Zhang, Wenhang Ge, Wei-Shi Zheng, Chunhua Shen
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 |
|---|---|---|---|
| Replace a single video representation with multiple mask-generated prototypes and text-adaptive matching (ablation) | Frozen on MSR-VTT · MSR-VTT | SumR: 167.9 -> 173.2 | Improved |
| Add variance loss to mask-based prototype matching (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 173.2 -> 176.1 | Improved |
| Use mask aggregation rather than part-based prototype production (ablation) | TVMM on MSR-VTT · MSR-VTT | SumR: 169.2 -> 173.2 | Improved |
| Use mask aggregation rather than a two-layer decoder for prototype production (ablation) | TVMM on MSR-VTT · MSR-VTT | SumR: 170.9 -> 173.2 | Improved |
| Use three extra prototypes instead of one (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 172.0 -> 176.1 | Improved |
| Use three extra prototypes instead of two (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 175.6 -> 176.1 | No clear effect |
| Increase the extra prototype count from three to four (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 176.1 -> 173.1 | Hurt |
| Increase the extra prototype count from three to ten (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 176.1 -> 168.9 | Hurt |
| Generate an extra prototype for every token instead of using three extra prototypes (ablation) | TVMM-Mask on MSR-VTT · MSR-VTT | SumR: 176.1 -> 163.8 | Hurt |
| Add multiple prototype matching and variance loss (ablation) | Frozen baseline with one input frame · MSR-VTT, 1 frame | SumR: 133.3 -> 137.3 | Improved |
| Add multiple prototype matching and variance loss (ablation) | Frozen baseline with four input frames · MSR-VTT, 4 frames | SumR: 163.7 -> 170.4 | Improved |
| Add multiple prototype matching and variance loss (ablation) | Frozen baseline with eight input frames · MSR-VTT, 8 frames | SumR: 167.9 -> 176.1 | Improved |
Closest papers by meaning
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · similarity 0.88)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.88)
- Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval (CVPR 2025 · similarity 0.87)
- Rethinking Noisy Video-Text Retrieval via Relation-aware Alignment (CVPR 2025 · similarity 0.87)
- DiscoVLA: Discrepancy Reduction in Vision, Language, and Alignment for Parameter-Efficient Video-Text Retrieval (CVPR 2025 · similarity 0.87)
- Ambiguity-Restrained Text-Video Representation Learning for Partially Relevant Video Retrieval (AAAI 2025 · similarity 0.87)
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.
