Object pop-up: Can we infer 3D objects and their poses from human interactions alone?

Input Point Cloud Result Figure 1 . Human-Centric prediction of Human-Object Interaction. We explore Human-Object interaction when only the cues of the human pose interacting with an unobserved object are available ('Input'). We propose a neural model that, for the first time, can infer the location of the object ('Result') from such input. This is possible even when our subject simulates the interactions with the object ('Reference Image').