Using AI to Age OSM Maps

The screenshot above, from the SynthMap Demo, shows a side-by-side view of an Open Street Map and an AI generated map of the same OSM data changed to look like a 19th Century era Ordnance Survey map. This Victorian cosplaying map (on the right) was developed by Zekun Li, of the University of Minnesota, who trained an AI to transform OpenStreetMap data into images that resemble the style of early British Ordnance Survey maps.

As you can see the results of this attempt to Generate synthetic historical maps from OSM map tiles is very impressive. But what is the point of an artificially aged digital map? I hear you ask! To which question there is a very good answer. These synthetically generated vintage maps are being created in order to teach AI models how to read historic maps.  

You may remember that in August of last year the David Rumsey Map Collection unveiled a new Text on Maps feature which allows you to search one of the world's largest collections of digitized maps by text. Using this new Text on Maps feature you can now search for any word in order to see where it appears on any of the collection's 57,000+ vintage maps. For example here are the results for searching 57,000 vintage maps for gold mine.

The reason that we are now able to search the David Rumsey Map Collection for individual words on vintage maps is because of huge advances in AI text recognition. Zekun Li's generated synthetic maps make AI models tasked with reading map labels even better. These AI generated maps closely resemble historical maps in style but have accurate 21st-century label information. Which means that the AI text recognition models can actually be trained on maps which have visual challenges similar to those in the actual vintage maps, but also have text labels which we know are accurate.This 'clean' data can then be used to train AI text recognition models on how to see and read map labels on real vintage maps. 

You can read more about Zekun Li's work on Maps with a sense of the past: what are synthetic maps, and why do we love them? on the National Library of Scotland blog.