Research map / Video / Video-language understanding
QueryStream: Advancing Streaming Video Understanding with Query-Aware Pruning and Proactive Response
Kairui Zhang, Zhenyu Yang, Bing Wang, Shengsheng Qian, Changsheng Xu
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 query-agnostic DTD with query-aware QDP at ~55% keep rate | TimeChat-Online-7B · StreamingBench | overall accuracy: 74.32 -> 75.32 | Improved |
| replace DTD with QDP at ~30% keep rate | TimeChat-Online-7B · StreamingBench | overall accuracy: 72.96 -> 74.04 | Improved |
| replace DTD with QDP | TimeChat-Online-7B · OVO-Bench | overall accuracy: 46.7 (full) -> 49.4 | Improved |
| QueryStream at 20.0% keep vs TimeChat-Online at 55.4% keep | TimeChat-Online-7B · OVO-Bench | overall accuracy: 47.6 -> 47.5 | Mixed |
| apply QDP zero-shot to Qwen2.5-VL (52.4% keep) | Qwen2.5-VL-7B · VideoMME | overall accuracy: 63.2 -> 63.6 | Improved |
| apply QDP to Qwen2.5-VL on long videos | Qwen2.5-VL-7B · VideoMME long | accuracy: 50.4 -> 52.6 | Improved |
| prune to 16.6% keep rate with QueryStream | TimeChat-Online-7B · LongVideoBench | accuracy: 57.3 -> 58.0 | Improved |
| add query-aware QDP at ~57% keep rate versus no pruning | TimeChat-Online-7B · StreamingBench | overall accuracy: 75.36 -> 75.32 | No clear effect |
| combine semantic and temporal filters rather than temporal filter alone (ablation) | QDP · StreamingBench | overall accuracy: 74.76 -> 75.32 | Improved |
| combine semantic and temporal filters rather than semantic filter alone (ablation) | QDP · StreamingBench | overall accuracy: 74.52 -> 75.32 | Improved |
| add density gate to relevance trigger (full RTAR) (ablation) | RTAR · OVO-Bench Forward Active Responding | timeliness score: 30.2 -> 34.6 | Improved |
| add relevance gate to density-only trigger (ablation) | RTAR · OVO-Bench Forward Active Responding | avg accuracy: 36.8 -> 40.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Dispider-7B — Dispider: Enabling Video LLMs with Active Real-Time Interaction via Disentangled Perception, Decision, and Reaction (CVPR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.90)
- LiveStar: Live Streaming Assistant for Real-World Online Video Understanding (NeurIPS 2025 · similarity 0.89)
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.89)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.89)
- SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding (ICLR 2025 · similarity 0.89)
- StreamBridge: Turning Your Offline Video Large Language Model into a Proactive Streaming Assistant (NeurIPS 2025 · similarity 0.88)
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.
