AI-Urban-Sketching: Deep Learning and Automating Design Perception for Creativity

Abstract

The paper reconsiders style transfer with generative adversarial networks (GANs) as a powerful means towards a machinic extraction of perception, one that learns how to imitate how a human might spatially abstract, translate and eventually create designs. The aim is to investigate the potential of deep learning a mapping between two domains, one being the perceived reality of an urban scene, and the other, its representation on a sketch. The creative discipline under consideration in this paper is that of architecture.