EgoObjects

EgoObjects是由Meta AI创建的大规模第一视角数据集，包含超过9000个视频，由来自50多个国家的250名参与者使用4种可穿戴设备收集。数据集涵盖368个对象类别，共有超过65万个对象标注，每个对象都带有实例级标识符。EgoObjects旨在捕捉相同对象在多样背景复杂性、周围对象、距离、光照和相机运动下的表现。该数据集通过多阶段联邦标注流程进行标注，以适应数据集的持续增长。EgoObjects适用于增强现实和机器人技术中的细粒度对象理解研究，特别是在需要区分同一类别中不同实例的应用场景中。

EgoObjects is a large-scale first-person perspective dataset developed by Meta AI. It contains over 9,000 videos, collected by 250 participants from more than 50 countries using four types of wearable devices. The dataset encompasses 368 object categories, with a total of more than 650,000 object annotations, and each object is assigned a unique instance-level identifier. EgoObjects is intended to capture the visual characteristics of the same object under diverse contextual complexities, surrounding objects, distances, lighting conditions, and camera motions. This dataset is annotated via a multi-stage federated annotation workflow to support its continuous growth. EgoObjects is suitable for fine-grained object understanding research in augmented reality and robotics, particularly in application scenarios that require differentiating between distinct instances within the same object category.