APVR: Hour-Level Long Video Understanding with Adaptive Pivot Visual Information Retrieval (2506.04953v2)

Abstract: Current multimodal LLMs (MLLMs) struggle with hour-level video understanding, facing significant challenges not only in modeling the substantial information volume of long videos but also in overcoming the memory wall and resource constraints during both training and inference. Although recent training-free approaches have alleviated resource demands by compressing visual features, their reliance on incomplete visual information limits the performance potential. To address these limitations, we propose \textbf{A}daptive \textbf{P}ivot \textbf{V}isual information \textbf{R}etrieval (\textbf{APVR}), a training-free framework that hierarchically retrieves and retains sufficient and important visual information. It breakthroughs the memory wall limitation via two complementary components: Pivot Frame Retrieval employs query expansion and iterative spatio-semantic confidence scoring to identify relevant video frames, and Pivot Token Retrieval performs query-aware attention-driven token selection within up to 1024 pivot frames. This dual granularity approach enables the processing of hour-long videos while maintaining semantic fidelity. Experimental validations demonstrate significant performance improvements, achieving 64.9\% on LongVideoBench and 68.4\% on VideoMME, which are state-of-the-art results for both training-free and training-based approaches. Meanwhile, our method provides plug-and-play integration capability with existing MLLM architectures.