Projecting emotions from artworks to maps using neural style transfer

Abstract. Recent advances in deep learning have facilitated the exchange of styles and textures between input images to create unique synthesised outputs. This paper assesses the applicability of neural style transfer to cartography and evaluates to what degree emotions attached to input images can be preserved in maps co-created by human and algorithm. As a source of emotions we utilized personal paintings created during a workshop with international artists at the School of Machines, Making & Make-Believe in August 2018. The neural style transfer was used as a tool to transfer the characteristics of the artworks onto the map. Differences in emotion perception between human-made textures and generated maps were assessed with an online survey completed by 1187 users. The results confirmed that emotional descriptions remain the same before and after the procedure of neural style transfer. The users perceived artificially generated maps as interesting and visually pleasing artefacts. Artworks with variety of line, point and surface depictions were the most suitable algorithm inputs and achieved better visual results in representing the map content. After analysing the neural style transfer technique and identifying its limitations for cartographic style and map content, we conclude with plausible directions for future research.