ego4d

EGO4D是世界上最大的以自我为中心的 (第一人称) 视频ML数据集和基准测试套件，在五个新的基准测试任务中拥有3,600小时 (和计数) 的密集叙述视频和广泛的注释。它涵盖了来自全球74个地点和9个不同国家的926名独特相机佩戴者在野外捕获的数百种日常生活活动场景 (家庭，户外，工作场所，休闲等)。视频的部分伴随有音频、环境的3D网格、眼睛注视、立体和/或来自同一事件处的多个自我中心相机的同步视频。数据收集方法旨在维护严格的隐私和道德标准，参与者同意并在相关情况下采用强大的去身份程序。

EGO4D is the world’s largest egocentric (first-person) video machine learning dataset and benchmark suite, boasting 3,600 hours (and counting) of densely narrated video and extensive annotations across five novel benchmark tasks. It covers hundreds of daily activity scenarios (home, outdoor, workplace, leisure, etc.) captured in the wild by 926 unique camera wearers from 74 locations across 9 distinct countries. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo footage and/or synchronized multi-egocentric camera footage from the same event. The data collection methodology is designed to uphold strict privacy and ethical standards, with participant consent and robust de-identification procedures implemented where applicable.