Deep generative modeling tasks: Automatic generation of building facades with Pix2Pix GAN for Hong Kong city expansion and renovation

Hong Kong's rapid urbanization and expansion have resulted in a high incidence of new urban construction projects as well as building renovation and restoration activities. The above architectural activities in the city are characterized by inconsistency in design styles and inconsistency with the neighborhood environment. The proposal of a self-weakening design style of different designers in the same neighborhood should be taken as a practical consideration at the early stage of design. In this research, a dataset of old building facades in Hong Kong is provided, and the method of training a deep convolutional neural network is used to realize the coupling of Pix2Pix GAN algorithm to the whole or local design generation of building facade in Hong Kong. Moreover, a trained network based on the architectural styles of Hong Kong is provided with 160 sets of collected original image datasets of Hong Kong building facades for organizing pre-calibration. It classifies the elemental information of complex building facades for later training of the network and automatic generation to give new construction and renovation schemes a replicable technical route.