Active Video Perception: Iterative Evidence Seeking for Agentic Long Video Understanding
This addresses the problem of inefficient and inaccurate video analysis for AI systems handling long videos, representing a novel method rather than an incremental improvement.
The paper tackles the challenge of long video understanding by proposing Active Video Perception (AVP), an evidence-seeking framework that actively decides what, when, and where to observe in videos, resulting in a 5.7% improvement in average accuracy while reducing inference time to 18.4% and input tokens to 12.4% compared to the best agentic method.
Long video understanding (LVU) is challenging because answering real-world queries often depends on sparse, temporally dispersed cues buried in hours of mostly redundant and irrelevant content. While agentic pipelines improve video reasoning capabilities, prevailing frameworks rely on a query-agnostic captioner to perceive video information, which wastes computation on irrelevant content and blurs fine-grained temporal and spatial information. Motivated by active perception theory, we argue that LVU agents should actively decide what, when, and where to observe, and continuously assess whether the current observation is sufficient to answer the query. We present Active Video Perception (AVP), an evidence-seeking framework that treats the video as an interactive environment and acquires compact, queryrelevant evidence directly from pixels. Concretely, AVP runs an iterative plan-observe-reflect process with MLLM agents. In each round, a planner proposes targeted video interactions, an observer executes them to extract time-stamped evidence, and a reflector evaluates the sufficiency of the evidence for the query, either halting with an answer or triggering further observation. Across five LVU benchmarks, AVP achieves highest performance with significant improvements. Notably, AVP outperforms the best agentic method by 5.7% in average accuracy while only requires 18.4% inference time and 12.4% input tokens.