Juntae Lee

Rethinking RAG in Long Videos: What to Retrieve and How to Use It?

This work introduces V-RAGBench, a benchmark of query, evidence chunk, answer$\rangle$ triplets over hour-scale videos, in which each answer depends on a unique evidence chunk, enabling decoupled evaluation of retrieval and generation, and CARVE, a training-free method that runs parallel retrievers across configuration...