Search papers, labs, and topics across Lattice.
This paper introduces Density-Aware Graph Construction (DAGC), a novel method that separates the temporal granularity of indexing from that of evidence in long-video retrieval-augmented generation (RAG). By creating a compact, density-adaptive graph index that merges visually redundant chunks, DAGC achieves significant efficiency gains, retaining only 40-50% of the original graph nodes while maintaining nearly 99% of the original QA performance. Experiments demonstrate that this approach accelerates processing by 1.3 to 1.7 times across various long-video models and pipelines, highlighting the potential for improved scalability in video understanding tasks.
Coarse indexing can dramatically speed up long-video retrieval without sacrificing answer quality, achieving up to 1.7x acceleration.
Graph-based retrieval-augmented generation (RAG) provides a scalable paradigm for long-video understanding, but existing systems typically inherit a fixed temporal granularity from video segmentation when constructing their retrieval index. We argue that this design unnecessarily couples indexing granularity with evidence granularity: coarse representations can often suffice for locating relevant temporal regions, while fine-grained evidence remains important for downstream reasoning. We propose \textbf{Density-Aware Graph Construction (DAGC)}, a training-free approach that decouples a query-independent coarse retrieval index from the original fine-grained evidence space. DAGC constructs a compact, density-adaptive graph index by merging visually redundant neighboring chunks, while preserving mappings to the original temporal units. Retrieved coarse regions are subsequently expanded back to the original chunk granularity for fine-grained evidence refinement and answer generation. Experiments on MLVU, VideoMME, and LongVideoBench show that DAGC retains only about 40--50\% of the original graph nodes and achieves $1.3$--$1.7\times$ end-to-end wall-clock acceleration while preserving approximately 99\% of the original QA performance. The gains transfer across different LVLM backbones and video RAG pipelines, suggesting that long-video RAG need not maintain the same temporal granularity for indexing and evidence reasoning.
