MasterplanGAN: Facilitating the smart rendering of urban master plans via generative adversarial networks

Abstract

This study proposes a prototype for the smart rendering of urban master plans via artificial intelligence algorithms, a process which is time-consuming and relies on professionals’ experience. With the help of crowdsourced data and generative adversarial networks (GAN), a generation model was trained to provide colorful rendering of master plans similar to those produced by experienced urban designers. Approximately 5000 master plans from Pinterest were processed and CycleGAN was applied as the core algorithm to build this model, the so-called MasterplanGAN. Using the uncolored input design files in an AutoCAD format, the MasterplanGAN can provide master plan renderings within a few seconds. The validation of the generated results was achieved using quantitative and qualitative judgments. The achievements of this study contribute to the development of automatic generation of previously subjective and experience-oriented processes, which can serve as a useful tool for urban designers and planners to save time in real projects. It also contributes to push the methodological boundaries of urban design by addressing urban design requirements with new urban data and new techniques. This initial exploration indicates that a large but clear picture of computational urban design can be presented, integrating scientific thinking, design, and computer techniques.