Visualizing Semiotics with Generative Adversarial Networks

Abstract

Designers use semiotics to create visual artifacts. The design for a flight attendant’s uniform may be modified to look more “alert,” less “austere,” or more “practical.” The form of a house may be modified to appear more “futuristic,” a car more “friendly,” and a pair of sneakers, “evil.” This process depends on a designer’s existing visual and conceptual vocabulary. This thesis develops a computational approach to augment a designer’s visual form-finding process and enables the designer to discover visual forms beyond her explicit visual and conceptual repertoire. It shows that visual semiotics can be learned by a neural network which can then be used to explore the latent space of a Generative Adversarial Network (GAN). Simultaneously it shows that exploring the latent space of a GAN using abstract language expands the range of visual possibilities