MatTrans: Material Reflectance Property Estimation of Complex Objects with Transformer

Material Reflectance Property Estimation of an object is challenging and it can be used in realistic rendering to make the appearance of objects realistic. Current research focuses primarily on the near-planar objects, with little attention paid to complex-shaped objects. In this paper, we propose a method called MatTrans to estimate geometry and material reflectance properties with Transformer. Specifically, a Transformer Encoder module is designed to fuse local and global information for each material property respectively, and then a cascaded network with residual learning is introduced to estimate the geometry and reflectance properties of any 3D object surface from a single image. Extensive experiments validate that our method brings a clear improvement over previous methods for single-shot capture of spatially varying BRDFs.