Research map / Video / Video-language understanding
ViSpeak: Visual Instruction Feedback in Streaming Videos
Shenghao Fu, Qize Yang, Yuan-Ming Li, Yi-Xing Peng, Kun-Yu Lin, Xihan Wei, Jian-Fang Hu, Xiaohua Xie, Wei-Shi Zheng
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 a trained informative head instead of the language-model head for proactive output (ablation) | ViSpeak · StreamingBench PO | PO score: 30.00 -> 34.80 | Improved |
| predict from the last visual token instead of the <seg> token (ablation) | ViSpeak informative head · StreamingBench PO | PO score: 34.80 -> 36.00 | Improved |
| jointly train the informative head with the LLM (ablation) | ViSpeak · StreamingBench PO | PO score: 36.00 -> 38.80 | Improved |
| jointly train the informative head with the LLM (ablation) | ViSpeak · StreamingBench overall | Overall: 60.91 -> 62.00 | Improved |
| combine the two input streams with different methods (Adaptive Sum, Linear, Add) (ablation) | ViSpeak template alignment · MME, MVBench, Video-MME | MVBench: 54.12 / 52.95 / 54.27 | No clear effect |
| template alignment to two-stream format versus original VITA 1.5 (ablation) | VITA 1.5 · MME | MME total: 2353.5 -> 2237.0 | Hurt |
| add offline video data (SMILE, IntentQA, Social-IQ) to ViSpeak-Instruct (ablation) | ViSpeak · ViSpeak-Bench | Overall: 2.70 -> 2.76 | Improved |
| stage 3 finetuning on ViSpeak-Instruct after stage 2 | ViSpeak · StreamingBench PO | PO score: 38.80 -> 50.80 | Improved |
| stage 3 finetuning on ViSpeak-Instruct | ViSpeak · StreamingBench real-time visual understanding | All score: 74.36 -> 70.44 | Hurt |
| give models explicit prompts about response type on ViSpeak-Bench | baseline MLLMs · ViSpeak-Bench | stability | 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 — 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
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.87)
- OmniMMI: A Comprehensive Multi-modal Interaction Benchmark in Streaming Video Contexts (CVPR 2025 · similarity 0.87)
- A Benchmark for Omni-Modal Reasoning in Long Videos (NeurIPS 2026 · similarity 0.87)
- StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos (NeurIPS 2026 · similarity 0.87)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.87)
- VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment (CVPR 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.
