Interactive Evolutionary Design Method of Tile Shape by GAN Reflecting Adjectives

A method of designing tile shapes using interactive evolutionary computing (IEC) with generative adversarial network (GAN) is proposed. Designing complex tile shape is difficult for ordinary people due to mathematical constraints and artistry requirements. The proposed method automatically generates shapes which reflect adjectives the user wants to express using GAN, and the shapes are refined by human interaction with IEC, so that the obtained shape becomes fully satisfactory to the user. Here, datasets of shapes are prepared for several adjectives first, and GAN is trained using random latent vectors to generate shapes for each adjective. The values in the latent vector are controlled by IEC to consider the user’s preference. The proposed method is shown to be effective in computer simulation and its subjective test.