GROVE: Growing and Reasoning over Temporally Stratified Memory from Streaming Video Experience
GROVE uses one causally built video memory for both recall and proactive assistance.
The paper describes a training-free framework for wearable assistants working from a continuous video stream. It stores evidence across time-stamped moments, episodes, and recurring cross-day patterns, with retrieval methods matched to each level. Reactive question answering and proactive help use the same memory interface, differing only in what starts retrieval. The authors report best results among compared methods on benchmarks including MM-lifelong and EgoServe. HF Daily Papers' note
The paper describes a training-free framework for wearable assistants working from a continuous video stream. It stores evidence across time-stamped moments, episodes, and recurring cross-day patterns, with retrieval methods matched to each level. Reactive question answering and proactive help use the same memory interface, differing only in what starts retrieval. The authors report best results among compared methods on benchmarks including MM-lifelong and EgoServe. HF Daily Papers' note
score 5