Research map / Video / Video-language understanding
Video Streaming Thinking: VideoLLMs Can Watch and Think Simultaneously
Yiran Guan, Liang Yin, Dingkang Liang, Jianzhong Ju, Zhenbo Luo, Jian Luan, Yuliang Liu, Xiang Bai
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 |
|---|---|---|---|
| mix 30K VST streaming-thought data with 20K LLaVA-Vid in SFT instead of 50K LLaVA-Vid (ablation) | Qwen2.5-VL-7B · OVO-Bench overall | accuracy: 52.3 -> 57.1 | Improved |
| add VST data to SFT mix (ablation) | Qwen2.5-VL-7B · VideoMME | accuracy: 61.8 -> 63.1 | No clear effect |
| add VST-SFT to base model (ablation) | Qwen2.5-VL-7B · OVO-Bench Backward | accuracy: 47.5 -> 56.7 | Improved |
| add VST-RL to base model (ablation) | Qwen2.5-VL-7B · OVO-Bench Forward | accuracy: 41.9 -> 54.6 | Improved |
| add RL on top of SFT (ablation) | VST-7B · OVO-Bench overall | accuracy: 57.4 -> 59.3 | Improved |
| add RL on top of SFT (ablation) | VST-7B · VideoMME | accuracy: 63.0 -> 64.9 | Improved |
| apply VST training to 3B base (ablation) | Qwen2.5-VL-3B · StreamingBench | accuracy: 67.8 -> 75.5 | Improved |
| apply VST training to 7B base (ablation) | Qwen2.5-VL-7B · StreamingBench | accuracy: 71.7 -> 79.5 | Improved |
| apply VST training to 32B base (ablation) | Qwen2.5-VL-32B · StreamingBench | accuracy: 71.5 -> 80.7 | Improved |
| apply VST training to 7B base (ablation) | Qwen2.5-VL-7B · VideoHolmes | accuracy: 32.9 -> 41.9 | Improved |
| increase max streaming thinking steps from 1 to 16 (ablation) | VST-7B · OVO-Bench Backward | accuracy: 53.3 -> 57.5 | Improved |
| increase max streaming thinking steps beyond 4 (ablation) | VST-7B · OVO-Bench Real-Time and Forward | accuracy | No clear effect |
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.
- Streamforest-7B — StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025) · better on 3 of 3 shared results
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
- Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models (CVPR 2026 · similarity 0.91)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.91)
- StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos (NeurIPS 2026 · similarity 0.90)
- SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding (ICLR 2025 · similarity 0.89)
- StreamOV: Streaming Omni-Video Understanding via Evidence-Guided Memory and Response Triggering (NeurIPS 2026 · similarity 0.89)
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · 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.
