HKUDS Releases VideoRAG for Local Search Across Hundreds of Hours of Video

HKUDS, the data intelligence research team at the University of Hong Kong, has released VideoRAG, an open-source retrieval-augmented generation framework that processes hundreds of hours of long-form video locally and answers questions in natural language, along with a desktop app built on it called Vimo Desktop. It is characterized by the ability to handle hundreds of hours of video on a single RTX 3090 (24GB) and by its cross-video contextual understanding.