We will showcase a large multiview dataset called HUMBI (HUman Multiview Behavioral Imaging), for human body expressions with natural clothing. The goal of HUMBI is to facilitate modeling view-specific appearance and geometry of gaze, face, hand, body, and garment from assorted people. HUMBI pushes to two extremes: views and subjects. The dataset is composed of 772 distinctive subjects with natural clothing across diverse age, gender, ethnicity, and physical condition captured by 107 HD synchronized cameras (68 cameras facing the frontal body). With the multiview image streams, we provide high fidelity 3D representations of body expressions using 3D mesh models for view-specific appearance and geometry. For the first time, we formulate a new benchmark challenge on pose-guided person multiview image generation.
Website: humbi-data.net
