HUMBI

HUMBI是一个大规模多视角人体表达数据集，由明尼苏达大学创建，旨在促进对人体主要信号（包括视线、面部、手部、身体和服装）的视角特定外观和几何建模。该数据集通过107个同步高清摄像机捕捉了772个不同性别、种族、年龄和风格的独特个体，使用3D网格模型重建身体表达的几何形状，允许表示视角特定的外观。HUMBI在学习和重建完整人体模型方面表现出高效性，并与现有有限视角和主题的人体表达数据集互补。数据集的应用领域包括扩展3D中多样化人体表达的逼真度，这是实现真实社交远程存在的关键因素。

HUMBI is a large-scale multi-view human expression dataset created by the University of Minnesota, aiming to facilitate view-specific appearance and geometric modeling of key human signals including gaze, face, hands, body and clothing. This dataset captures 772 unique individuals with diverse genders, ethnicities, ages and styles via 107 synchronized high-definition cameras, and reconstructs the geometry of body expressions using 3D mesh models, enabling representation of view-specific appearances. HUMBI demonstrates high efficiency in learning and reconstructing full-body human models, and complements existing human expression datasets with limited viewpoints and subjects. Application scenarios of this dataset include enhancing the realism of diverse human expressions in 3D, which is a critical factor for realizing realistic social telepresence.