RoBus: A Multimodal Dataset for Controllable Road Networks and Building Layouts Generation

TL;DR

RoBus addresses the lack of large-scale, multimodal data for controllable city-scale generation of road networks and building layouts. The authors introduce RoBus, a multimodal dataset consisting of images, topological graphics, and descriptive texts, with 72,400 paired samples across roughly 80,000 km^2, designed to support geometry-constrained and property-conditioned generation. They establish a comprehensive benchmark with metrics for quality, diversity, validity, and urban properties, and propose baselines that inject urban characteristics into generation of roads and vectored buildings. They further demonstrate practical applicability by exporting outputs to OpenDRIVE and integrating with driving simulators and game engines, highlighting immediate utility for urban design, autonomous driving simulations, and urban-focused gaming.

Abstract

Automated 3D city generation, focusing on road networks and building layouts, is in high demand for applications in urban design, multimedia games and autonomous driving simulations. The surge of generative AI facilitates designing city layouts based on deep learning models. However, the lack of high-quality datasets and benchmarks hinders the progress of these data-driven methods in generating road networks and building layouts. Furthermore, few studies consider urban characteristics, which generally take graphics as analysis objects and are crucial for practical applications, to control the generative process. To alleviate these problems, we introduce a multimodal dataset with accompanying evaluation metrics for controllable generation of Road networks and Building layouts (RoBus), which is the first and largest open-source dataset in city generation so far. RoBus dataset is formatted as images, graphics and texts, with $72,400$ paired samples that cover around $80,000km^2$ globally. We analyze the RoBus dataset statistically and validate the effectiveness against existing road networks and building layouts generation methods. Additionally, we design new baselines that incorporate urban characteristics, such as road orientation and building density, in the process of generating road networks and building layouts using the RoBus dataset, enhancing the practicality of automated urban design. The RoBus dataset and related codes are published at https://github.com/tourlics/RoBus_Dataset.

Figures (11)

• Figure 1: An example from the RoBus Dataset, including images, graphics, and texts that describe road networks and building layouts in a city tile. The RoBus Dataset is scalable for various 3D urban generation tasks and practical applications.
• Figure 2: Considering urban characteristics, we classify buildings in a city block according to density and average height.
• Figure 3: The Generating Pipeline of RoBus Dataset.
• Figure 4: Visualization of Statics of RoBus Dataset.
• Figure 5: Framework of Task III.