: A Meta AI paper that uses similar large-scale video datasets to train AI models to "understand" physical world interactions without explicit labels.
: Unlike general video datasets, this focuses on skilled tasks like cooking, dancing, music, and sports, where precise body movements and tool interactions are key [2]. 2023-01-19-00-30-09.mp4
: It captures the same activity from both the participant's wearable camera and surrounding static cameras, allowing AI to learn how first-person views relate to the broader environment [1]. : A Meta AI paper that uses similar
: The predecessor to Ego-Exo4D, focusing purely on first-person "daily life" videos. 2023-01-19-00-30-09.mp4
If you are interested in how this specific type of video data is used, these follow-up papers are also highly relevant: