Transferring multiscale map styles using generative adversarial networks

Abstract

The advancement of the Artificial Intelligence (AI) technologies makes it\npossible to learn stylistic design criteria from existing maps or other visual\nart and transfer these styles to make new digital maps. In this paper, we\npropose a novel framework using AI for map style transfer applicable across\nmultiple map scales. Specifically, we identify and transfer the stylistic\nelements from a target group of visual examples, including Google Maps,\nOpenStreetMap, and artistic paintings, to unstylized GIS vector data through\ntwo generative adversarial network (GAN) models. We then train a binary\nclassifier based on a deep convolutional neural network to evaluate whether the\ntransfer styled map images preserve the original map design characteristics.\nOur experiment results show that GANs have a great potential for multiscale map\nstyle transferring, but many challenges remain requiring future research.\n