CVPR2024

Intrinsic Image Diffusion for Indoor Single-view Material Estimation

Peter Kocsis, Vincent Sitzmann, Matthias Nießner

Abstract

peter-kocsis.github.io/IntrinsicImageDiffusion/ Figure 1 . Intrinsic Image Diffusion. We present Intrinsic Image Diffusion for single-view material estimation of indoor scenes. Since appearance decomposition is a highly ambiguous task, we propose to use a conditional generative model to predict multiple solutions and utilize the strong prior of recent diffusion models [36] . Our approach gives detailed and consistent material estimations on complex indoor scenes, outperforming recent state-of-the-art methods and allows for high-quality controllable lighting optimization.