UnrealEgo

UnrealEgo是由马克斯·普朗克信息学研究所和庆应大学创建的大规模自然主义数据集，专为第一人称3D人体姿态估计设计。该数据集基于先进的概念，即配备两个鱼眼摄像头的眼镜，可在不受限制的环境中使用。UnrealEgo包含45万张野外立体图像（总计90万张图像），拥有现有第一人称数据集中最大的动作种类。数据集创建过程中，使用Unreal Engine设计虚拟原型，并将其附加到3D人体模型上以进行立体视图捕捉。UnrealEgo的应用领域包括XR技术和运动与健康分析，旨在解决在不受限制的环境中捕捉日常人体活动的问题。

UnrealEgo is a large-scale naturalistic dataset developed by the Max Planck Institute for Informatics and Keio University, specifically tailored for first-person 3D human pose estimation. Built on an advanced concept of eyeglasses equipped with two fisheye cameras, this dataset supports deployment in unconstrained environments. UnrealEgo contains 450,000 stereo images captured in the wild, totaling 900,000 images overall, and features the most diverse set of action categories among existing first-person datasets. During the dataset construction phase, Unreal Engine was employed to design virtual prototypes, which were then attached to 3D human body models for stereo view capture. The application domains of UnrealEgo include XR technologies, motion and health analysis, and it aims to address the challenge of capturing daily human activities in unconstrained environments.