A Data-Driven Method for Intrinsic Decomposition of 3D City Reconstruction Scene

Nowadays, 3D reconstruction pipelines based on structure-from motion and multi-view stereo techniques can reconstruct meticulous, large-scale geometry models and high-resolution textures. However, due to the shadow parts of the acquired images, the reconstructed texture map has dark areas, which is not convenient for relighting, which limits the use of the reconstruction model in VR/AR programs or games. In this paper, we introduce a large-scale synthetic city intrinsic images dataset for evaluating intrinsic decomposition of city scenes. Given our database, we developed a data-driven method for obtaining the diffuse albedo of reconstructed scenes, our method achieves good results in our intrinsic images dataset and real images.