1 paper
Zhe Jin, Zhimin Lin, Bin Zheng +2
Graph-based retrieval-augmented generation (RAG) provides a scalable paradigm for long-video understanding, but existing systems typically inherit a fixed temporal granularity from…