Research map / Video / Video-language understanding
Symphony: A Cognitively-Inspired Multi-Agent System for Long-Video Understanding
Haiyang Yan, Hongyun Zhou, Peng Xu, Xiaoxue Feng, Mengyi Liu
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 |
|---|---|---|---|
| add a separate reflection agent instead of planner self-reflection (ablation) | Symphony · LVBench | accuracy (%): -2.5% when removed | Improved |
| add a subtitle agent instead of feeding full subtitles to the planner (ablation) | Symphony · LVBench | accuracy (%): -1.4% when removed | Improved |
| add a specialized visual perception agent instead of merging its tools into the planner (ablation) | Symphony · LVBench | accuracy (%): +2.2% | Improved |
| use the full multi-agent system instead of the planning agent alone (ablation) | Symphony · LVBench | accuracy (%): 65.7 -> 71.8 | Improved |
| use VLM-based scoring grounding with CLIP fallback instead of CLIP-only retrieval (ablation) | Grounding agent · LVBench | accuracy (%): 52.2 -> 71.8 | Improved |
| always use VLM scoring (no CLIP) instead of adaptive selection (ablation) | Grounding agent · LVBench | accuracy (%): 71.8 -> 72.1, time 54.8 -> 68.6 s | Mixed |
| use Seed 1.6VL instead of Qwen2.5VL-7B as scoring VLM (ablation) | Grounding agent · LVBench | accuracy (%): 68.6 -> 71.8 | Improved |
| lower sampling FPS from 0.5 to 0.25 (ablation) | Grounding agent · LVBench EU subset | accuracy (%): 68.6 -> 62.1 | Hurt |
| extend clip interval from 60 s to 120 s (ablation) | Grounding agent · LVBench EU subset | accuracy (%): 68.6 -> 64.7 | Hurt |
| increase FPS to 1 or shorten clip to 30 s (ablation) | Grounding agent · LVBench EU subset | accuracy (%): 68.6 -> 70.9 (FPS 1), 70.3 (30 s clip) | No clear effect |
| add 3-instance majority voting (ablation) | Symphony · LVBench | accuracy (%): 71.8 -> 73.7 | Improved |
| add 3-instance majority voting (ablation) | Symphony · MLVU | accuracy (%): 81.0 -> 83.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- LvAgent — LVAgent: Long Video Understanding by Multi-Round Dynamical Collaboration of MLLM Agents (ICCV 2025) · better on 1 of 2 shared results
- VideoChatA1 — VideoChat-A1: Thinking with Long Videos by Chain-of-Shot Reasoning (AAAI 2026) · 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
- LVAgent: Long Video Understanding by Multi-Round Dynamical Collaboration of MLLM Agents (ICCV 2025 · similarity 0.90)
- Think, Then Verify: A Hypothesis-Verification Multi-Agent Framework for Long Video Understanding (CVPR 2026 · similarity 0.88)
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.87)
- Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding (NeurIPS 2025 · similarity 0.87)
- Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation (CVPR 2025 · similarity 0.86)
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 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.
