RACER
A training-free reflective agent framework designed for keyframe selection in long video understanding. It separates query comprehension and tool retrieval into collaborative roles and incorporates a reflection loop to improve frame selection accuracy, without requiring additional model training.
Timeline (last 90 days)
The full name of RACER is Reflective Agent Coupling Query Interpretation and Tool-Based Retrieval.
RACER is a training-free reflective agent framework for key frame selection in long videos that can be deployed without any additional model training.
RACER decomposes the frame selection task into two steps: a lightweight Vid-LLM is responsible for rewriting complex queries into a series of sub-queries, and an embedding model serves as a retrieval tool to locate relevant evidence frames.
RACER's reflective loop design draws on self-examination mechanisms from frameworks such as ReAct and Reflexion.
RACER consistently improves long video understanding performance across multiple benchmarks.
Even when using components with limited capability, RACER can still achieve effective frame selection, and weaker components can synergistically enhance more powerful Vid-LLMs.
The RACER-related research was published as a paper on arXiv.
All Facts (7)
The full name of RACER is Reflective Agent Coupling Query Interpretation and Tool-Based Retrieval.
50%UnverifiedRACER is a training-free reflective agent framework for key frame selection in long videos that can be deployed without any additional model training.
50%UnverifiedRACER decomposes the frame selection task into two steps: a lightweight Vid-LLM is responsible for rewriting complex queries into a series of sub-queries, and an embedding model serves as a retrieval tool to locate relevant evidence frames.
50%UnverifiedRACER's reflective loop design draws on self-examination mechanisms from frameworks such as ReAct and Reflexion.
50%UnverifiedRACER consistently improves long video understanding performance across multiple benchmarks.
50%UnverifiedEven when using components with limited capability, RACER can still achieve effective frame selection, and weaker components can synergistically enhance more powerful Vid-LLMs.
50%UnverifiedThe RACER-related research was published as a paper on arXiv.
50%