Long-Video QA
QA: Question Answering
Overview
Long-video question answering studies how to answer questions that depend on events, people, or details spread across lengthy recordings. Common approaches select a small set of relevant frames, retrieve clips at multiple time scales, or organize content into event, entity, and scene-graph memories; agentic systems can search, verify evidence, and decide when to stop. Methods are evaluated on benchmarks such as Video-MME, LongVideoBench, LVBench, and egocentric video QA tasks, with growing attention to streaming and audio-visual settings. A central challenge is balancing accuracy and efficiency while ensuring answers are supported by correctly localized evidence, especially when clues are distant, personal, or spread across multiple scenes.
Latest papers
89 papers11 papers in the last four weeks, level with the four weeks before. 0.1% of all new papers.