Egocentric Video QA
QA: Question Answering
Overview
Egocentric video question answering asks models to answer questions from first-person footage, supporting applications such as activity understanding and personal assistants. Common approaches combine video-language models with frame or clip retrieval and structured memories such as captions and temporal scene graphs, then reason over spatial, temporal, and procedural evidence. Benchmarks test long-term recall, streaming anticipation, out-of-view object tracking, personalized queries, and performance across multiple viewpoints or domains. Models handle many local visual questions but still lose evidence across long timelines, confuse camera motion with changes in the scene, and answer confidently without support; improving grounding and memory while controlling latency and storage remains an open challenge.
Latest papers
46 papers7 papers in the last four weeks, up 75% on the four weeks before. 0.0% of all new papers.