Human-centric metrics for indoor scene assessment and synthesis

In recent years, many approaches have been proposed to analyze the affordance of 3D objects and to facilitate the synthesis of aesthetic and realistic indoor scenes. However, how to assess the quality of such synthesized 3D scenes is still a challenging problem. To address this, we present a novel approach through so-called Human-Centric Metrics (HCMs) for quantitatively evaluating the layout quality of certain objects, and thus to facilitate indoor scene synthesis. Our HCMs consider both the human-object factors to assess the functional accessibility of indoor objects, and the human-environment factors to assess the subjective comfort in the scene. Given a 3D scene with arranged furniture objects, our method automatically places human agents and then use HCMs to assess both the object arrangement and indoor environment based on the position/direction of a certain human agent. The conducted user study shows that the assessment capability of HCMs well conforms to the interior design knowledge. We also use the HCMs based assessment results to synthesize 3D indoor scenes by suggesting indoor objects and their layout given an empty room. A series of experimental results and comparisons are presented to validate the usability and effectiveness of our HCMs for virtual scene assessment and synthesis.