Material Reflectance Property Estimation of Complex Objects Using an Attention Network

Abstract

Material reflectance property modeling can be used in realistic ren-dering to generate realistic appearances for virtual objects. However, current works mainly focus on near plane objects. In this paper, we propose an end-to-end network framework with attention mecha-nism to estimate the reflectance properties of any 3D object surface from a single image, where a single attention module is used for each reflectance property respectively to learn the property specific features. We also generate a material dataset by rendering a set of 3D complex shape models. The dataset is suitable for reflectance property estimation of arbitrary complex shape objects. Experiments validate the proposed method.