Pose2Room: Understanding 3D Scenes from Human Activities

Abstract

2 SRIBD, CUHKSZ Figure 1. From an observed pose trajectory of a person performing daily activities in an indoor scene (left), we learn to estimate likely object configurations of the scene underlying these interactions, as set of object class labels and oriented 3D bounding boxes (middle). By sampling from our probabilistic decoder, we synthesize multiple plausible object arrangements (right). (Scene geometry is shown only for visualization.)