Learning optimal lighting patterns for efficient SVBRDF acquisition

Abstract

Digitally acquiring high-quality material appearance from the real-world is challenging, with applications in visual effects, e-commerce and entertainment. One popular class of existing work is based on hand-derived illumination multiplexing [Ghosh et al. 2009], using hundreds of patterns in the most general case [Chen et al. 2014]. We propose a novel framework [Kang et al. 2018] that automatically learns the lighting patterns for efficient reflectance acquisition, as well as how to faithfully reconstruct spatially varying anisotropic BRDFs and local frames from measurements under such patterns. Our core is an asymmetric deep autoencoder, consisting of a nonnegative, linear encoder which corresponds to the lighting patterns used in physical acquisition, and a stacked, nonlinear decoder which computationally recovers the BRDF information from photographs. We capture high-quality SVBRDFs with only 16 ∼ 32 lighting patterns, in 12 ∼ 25 seconds. Our framework is useful for increasing the efficiency in both novel and existing acquisition setups.