GENERATIVE LARGE-SCALE URBAN LAYOUT ANALYSIS AND SYNTHESIS

A building layout consists of a set of buildings in city blocks defined by a network of roads. Modeling and generating large-scale urban building layouts is of significant interest in computer vision, computer graphics, and urban applications. Researchers seek to obtain building features (e.g. building shapes, counts, and areas) at large scales. However, data quality and data equality challenge the generation and extraction of building features. Blurriness, occlusions, and noise from prevailing satellite images severely hinders performance of image segmentation, super-resolution, or deep-learning based translation networks. Moreover, large-scale urban layout generation struggles with complex and arbitrary shapes of building layouts, and context-sensitive nature of the city morphology, which prior approaches have not considered. Facing the challenges of data quality, generation robustness, and context-sensitivity of urban layout generation, In this thesis, we first address the data quality problem by combing globally-available satellite images and spatial geometric feature datasets, in order to create a generative modeling framework that enables obtaining significantly improved accuracy in per-building feature estimation as well as generation of visually plausible building footprints. Secondly, for generation robustness, We observe that building layouts are discrete structures, consisting of multiple rows of buildings of various shapes, and are amenable to skeletonization for mapping arbitrary city block shapes to a canonical form. In that, we propose a fully automatic approach to building layout generation using graph attention networks. The method generates realistic urban layouts given arbitrary road networks, and enables conditional generation based on learned priors. Nevertheless, we propose the approach addresses context-sensitivity by leveraging a canonical graph representation for the entire city, which facilitates scalability and captures the multi-layer semantics inherent in urban layouts. We introduce a novel graph-based masked autoencoder (GMAE) for city-scale urban layout generation. The method encodes attributed buildings, city blocks, communities and cities into a unified graph structure, enabling self-supervised masked training for graph autoencoder. Additionally, we employ scheduled iterative sampling for 2.5D layout generation, prioritizing the generation of important city blocks and buildings. Our method has proven its robustness by large-scale prototypical experiments covering heterogeneous scenarios from dense urban to sparse rural. It achieves good realism, semantic consistency, and correctness across the heterogeneous urban styles in 330 US cities.