Research map / Video / Video-language understanding
TVR: A Large-Scale Dataset for Video-Subtitle Moment Retrieval
Jie Lei, Licheng Yu, Tamara L. Berg, Mohit Bansal
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 XML's late-fusion ConvSE design instead of retrieval+re-ranking with early-fusion ExCL | XML · TVR test-public VCMR | R@1 IoU=0.7: 0.33 -> 3.25 | Improved |
| add ConvSE start-end detector instead of sliding-window proposals (ablation) | XML · TVR test-public VCMR | R@1 IoU=0.7: 1.91 -> 3.25 | Improved |
| add hierarchical video-then-moment retrieval (reduce candidates) to proposal-based MCN/CAL via MEE retrieval | MCN / CAL · TVR test-public VCMR | R@1 IoU=0.7: 0.04 -> 0.39 | Improved |
| use ConvSE kernel size 1 instead of 3, 5 or 7 (ablation) | ConvSE · TVR VCMR | performance | Hurt |
| add transformer self-encoder instead of bidirectional LSTM (ablation) | XML · TVR val VCMR, IoU=0.7 | R@1: 2.12 -> 2.62 | Improved |
| add transformer self-encoder instead of CNN encoder (ablation) | XML · TVR val VCMR, IoU=0.7 | R@1: 2.45 -> 2.62 | Improved |
| add modular (two) queries instead of one max-pooled query (ablation) | XML · TVR val VCMR, IoU=0.7 | R@1: 2.46 -> 2.62 | Improved |
| add I3D action features to ResNet appearance features (ablation) | XML · TVR val VCMR, IoU=0.7 | R@1: 2.28 -> 2.62 | Improved |
| add ResNet appearance features to I3D action features (ablation) | XML · TVR val VCMR, IoU=0.7 | R@1: 2.22 -> 2.62 | Improved |
| use both video and subtitle inputs instead of a single modality (ablation) | XML · TVR val VCMR | overall performance | Improved |
| increase retrieved videos from 10 to 200 (ablation) | XML · TVR VCMR IoU=0.5 | R@100: 31.05 -> 36.20 | Mixed |
| add Temporal Endpoint Feature (TEF) to MCN and CAL (ablation) | MCN / CAL / MEE+CAL · TVR test-public VCMR, IoU=0.5 | R@1: 0.97 -> 1.23 | Mixed |
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)
- Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers (CVPR 2024 · similarity 0.87)
- VERIFIED: A Video Corpus Moment Retrieval Benchmark for Fine-Grained Video Understanding (NeurIPS 2024 · similarity 0.86)
- Fewer Steps, Better Performance: Efficient Cross-Modal Clip Trimming for Video Moment Retrieval Using Language (AAAI 2024 · similarity 0.86)
- Generative Video Diffusion for Unseen Novel Semantic Video Moment Retrieval (AAAI 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.
