Symphony: A Cognitively-Inspired Multi-Agent System for Long-Video Understanding

Dimensions overall score 9.0

• Symphony achieves state-of-the-art performance on LVBench, LongVideoBench, VideoMME, and MLVU benchmarks.result 95%
• Symphony improves over the prior state-of-the-art method on LVBench by 5.0%.result 95%
• Symphony emulates human cognition patterns to decompose long-video understanding into fine-grained subtasks.method 90%
• Symphony incorporates a deep reasoning collaboration mechanism enhanced by reflection to improve reasoning capability.method 90%
• Symphony provides a VLM-based grounding approach to analyze LVU tasks and assess video segment relevance.technical 90%
• Symphony's VLM-based grounding approach enhances the ability to locate complex problems with implicit intentions and large temporal spans.technical 85%
• Simple task decomposition and collaboration mechanisms are insufficient for long-chain reasoning tasks in long-video understanding.limitation 80%
• Directly reducing time context through embedding-based retrieval may lose key information of complex problems in long-video understanding.limitation 80%
• Symphony achieves state-of-the-art performance on LVBench with a 5.0% improvement over the prior state-of-the-art method.result 95%
• Symphony's deep reasoning collaboration mechanism enhanced by reflection effectively improves reasoning capability for long-video understanding.method 85%
• Simple task decomposition and collaboration mechanisms are insufficient for long-chain reasoning tasks in long-video understanding.limitation 80%
• Directly reducing the time context through embedding-based retrieval may lose key information of complex problems in long-video understanding.limitation 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card