Transferring multiscale map styles using generative adversarial networks

ABSTRACT

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

RÉSUMÉ

Les avancées en intelligence artificielle (IA) permettent d’apprendre à une machine les critères de conception cartographique. Dans ce papier, nous proposons un nouveau cadre logiciel pour transférer les styles cartographiques. Les styles cartographiques dédiés aux cartographies telles que celles de Google Maps, OpenStreetMap ou même les styles utilisés par les artistes peintres peuvent être appris et transférés à des données SIG vectorielles grâce à deux réseaux antagonistes génératifs (GANs). Un classifieur binaire basé sur un réseau de neurones convolutif profond est entrainé pour évaluer si les images des styles transférés préservent les caractéristiques cartographiques. Nos résultats expérimentaux montrent que les GANs ont un fort potentiel pour le transfert des styles cartographiques mais qu’il reste de nombreux défis à relever.

KEYWORDS:

• GeoAI
• generative adversarial network
• style transfer
• convolutional neural network
• map design

Acknowledgments

The authors would like to thank Bo Peng at the University of Wisconsin-Madison, Fan Zhang from the MIT Senseable city lab, and Di Zhu from the Peking University for their helpful discussions for the research.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Yuhao Kang is a Master/Ph.D student at the Department of Geography, University of Wisconsin-Madison. He holds a bachelor’s degree in Geographic Information Science at the Wuhan University. His main research interests include Place-Based GIS, Spatio-temporal Data Mining, GeoAI, Human Mobility, and Urban Computing.

Dr. Song Gao is an Assistant Professor in GIScience at the Department of Geography, University of Wisconsin-Madison, where he leads the Geospatial Data Science Lab. He holds a Ph.D. in Geography at the University of California, Santa Barbara. His main research interests include Place-Based GIS, Geospatial Big Data Analytics, GeoAI, Human Mobility, and Urban Computing. He currently serves as the Associate Editor for Annals of GIS and the Editorial Board Member of PLOS ONE.

Dr. Robert E. Roth is the Faculty Director of the University of Wisconsin Cartography Lab and an Associate Professor in the University of Wisconsin-Madison Department of Geography. His research focuses on interactive, online, and mobile map design and visualization. He currently serves as the Vice Chair of the ICA Commission on Use, Users, and Usability and the Section Editor for Cartography and Visualization in the Geographic Information Science and Technology Body of Knowledge.

ORCID

Yuhao Kang http://orcid.org/0000-0003-3810-9450

Song Gao http://orcid.org/0000-0003-4359-6302

Robert E. Roth http://orcid.org/0000-0003-1241-318X

Notes

1 https://carto.com/developers/styling/cartocss/

2 https://www.mapbox.com/designer-maps/

3 https://tilemill-project.github.io/tilemill/

4 http://tilecache.org/

5 https://www.google.com/maps

6 https://www.openstreetmap.org

7 http://geoserver.org/

8 https://www.flickr.com/services/api/flickr.photos.search.html

9 https://www.geofabrik.de/data/download.html

Additional information

Funding

This research was funded by the Wisconsin Alumni Research Foundation and the Trewartha Graduate Research fund.