Generation of product design using GAN based on customer's kansei evaluation

Abstract

In recent years, deep learning has attracted much attention and various techniques have been proposed. GAN (Generative adversarial networks) is one such method. GAN uses images as the training set and learns to generate new images that are indistinguishable from the training set. In this study, A GAN-based design method that generates new products from the images of the customer's favorite products is proposed. The product images that customers evaluated as preferable are used as the training set of GAN. If the GAN fulfills its capabilities properly, the images generated from a customer's favorite product are more likely to be preferred by the customer. In the case study, the proposed method was applied to chair design. The generated chair images were first evaluated in terms of image quality, and then evaluated by subjects.