Architectural Facade Recognition and Generation through Generative Adversarial Networks
Yu Qiu, Jamal Malaeb, MA Wen-jun
Abstract
With the development of artificial intelligence technology, the ideas of machine learning have been introduced into the field of design in recent years. The research methods of “AI + Architecture” have brought new ideas for solving traditional problems. Generative Adversarial Network (GAN) is a machine learning model for image generation. Pix2pix is an improved version of GAN, which is specially designed to learn and generate pairs of image data with similar characteristics. In this study, Pix2pix is applied to the recognition and generation of building facade. The purpose is to explore the feasibility of using image generation technology to achieve rapid recognition and generation of building facade based on pix2pix. This paper also discusses the application scenarios of this technology. The existing building façade datasets and the self-made Chinese traditional building datasets are used to test and verify that pix2pix under different types of datasets can nicely identify and generate facade images. Then we summarize a set of working methods based on GAN to realize the overall or local reconstruction design of the facade, so as to provide new ideas for the improvement of the efficiency of related industries and the expansion of teaching tools.