State of the Map 2017: An Evolving Map Ecosystem
State of the Map 2017 highlighted the mutual dependence of individuals, NGOs, universities, and companies when it comes to the future of OpenStreetMap. This post outlines why this is a positive development and covers some of the takeaways from this year's international conference.
Getting to Aizuwakamatsu is quite a trek. From the impressive Narita Airport it takes about 3 trains and 4–5 hours. One can’t complain though. Hurtling through Japan at 300 km/h in comfort is not exactly an arduous experience. Tray table down, sushi box out, browsing through the program for State of the Map 2017.
For the uninitiated, State of the Map (SotM) is an annual gathering for OpenStreetMap, the crowdsourced map of our world. This was the international SotM, but other annual conferences take place in various countries and regions around the world. Aizuwakamatsu, Japan, was selected as the venue for the international 2017 gathering, in part due to the impressive efforts of mappers in Fukushima to grow the community and map the prefecture.
I won’t give the minute by minute lowdown on SotM 2017, but I will share some of the main takeaways from my favourite SotM thus far. This is by no means an exhaustive list, so make sure to check out the video recordings of the sessions.
Tsuruga-jo Castle in Aizuwakamatsu (photo by OSM Fukushima CC-BY-SA 3.0)
The OpenStreetMap ecosystem
This conference reinforced how diverse the OpenStreetMap ecosystem has become. What started out as a project driven by individual contributors, has evolved into a project that now includes not only these original contributors but NGOs, schools/universities, and companies.
This demonstrates the trust these institutions have in OpenStreetMap and the continued growth the project will hopefully enjoy. Each of these stakeholders has a mutual interest in contributing to the map, making it more accurate, and ensuring it is useful for as many people as possible.
NGOs have helped bring OpenStreetMap to the areas that need maps most, often facing incredible technical and bureaucratic challenges in doing so. Schools and universities have introduced a new generation to mapping, while providing valuable research into volunteered geographic information. Companies have brought a focus to OpenStreetMap, placing a greater emphasis on quality and using their resources to develop more efficient mapping tools.
And of course, we can’t forget the individuals who have and will continue to play a key role in OpenStreetMap. It is these individuals who spend countless hours of their own time maintaining repositories, taking on leadership roles, encouraging participation in the project, and editing the map. Together, these stakeholders are creating a map that is relied upon by millions, if not billions.
Mapbox continues to be an innovator in OpenStreetMap, coming up with new ways to improve the quality of data being added. Arun, Lily, Oindrila, and Manohar all gave insightful presentations on Mapbox’s approach, but here are two of the tools that caught my eye.
OSMCha with filters for Mapillary applied
I’d heard a bit about OSMCha prior to SotM and had played around with it briefly, but it was really worthwhile getting a demo from Manohar. OSMCha is a powerful tool to search changesets, with features that allow you to drill down to certain types of edits.
For Mapbox, this enables them to quickly see questionable edits and/or situations where an inappropriate source may have been used. On our side, it’s a useful tool for us to see how Mapillary imagery is being used in changesets across the world. The bounding box can be used to focus on a specific area, while comments and source can be searched for keywords such as
street-level. We can also see how iD and JOSM are being favoured by editors.
Navigation Data Map showing traffic signs Mapillary has detected in New York
Mapbox’s Navigation Data Map is an editing tool that highlights traffic signage that Mapillary has identified in an area. Using our vector tile layer and Mapbox styling, the Navigation Data Map allows an editor to quickly add turn restrictions, traffic signals, and speed limits to the map. The tool has been around for about a year, but is continually maintained and updated. You can clone the repo here.
Facebook, satellite imagery, and artificial intelligence
Drishtie Patel outlining the artificial intelligence methods Facebook is using to make map edits
Drishtie Patel from Facebook gave an excellent presentation on Facebook’s OpenStreetMap work in Thailand. You can also find an overview here. Since 2016, they’ve been experimenting with satellite imagery, using computer vision and deep neural nets to trace road networks, and from that create ways in OpenStreetMap.
When roads have been identified, they go through post processing before being loaded into Facebook’s fork of iD Editor. What I found particularly impressive is how Facebook is partnering with local mappers to validate what AI has detected. Local members are able to add more detail to the road edits, using their knowledge to specify the type of road that has been identified.
This kind of collaboration is very important because it has helped to add a lot of roads that were easy for individual contributors to miss. Both Arun from Mapbox and Drishtie from Facebook have pointed out the challenges of adding new areas to the map. It’s easier and more apparent to add the major roads and buildings, but the finer level of detail is often missed. It’s harder to pick up the little residential and rural roads that together make up the road network. Often these are the most important to add to the map, with many houses being connected to the map for the first time.
Commercial usage of OpenStreetMap
While I’m a great fan of OpenStreetMap, I have been guilty of opening Google Maps when I want to find the best public transport route or the nearest ramen restaurant. This is partly because the cartographers and designers at Google have done a superb job highlighting the key features you’d like to see on the map.
Historically OpenStreetMap was rather clunky and best for those with more patience than I. Thankfully useful apps like MAPS.ME & OSM.And have emerged. These apps use OpenStreetMap as a base map, but present it in an aesthetically appealing and more efficient way. They also allow you to download regions for offline use, an invaluable feature when you’re travelling.
In addition to these apps, other apps that we use every day are leveraging OpenStreetMap data, and often you wouldn’t even know it. Some of these rely purely on OpenStreetMap data, while others use OpenStreetMap as one of many sources to create the map that best fits their needs.
Take Snap for example which recently introduced Snap Map. You can use Snap Map to see a heat map of snaps by other users and see what’s going on nearby. You can imagine that this would become a more targeted way for Snap to further monetize their user base through events and local businesses. Apple also uses OpenStreetMap data, combining it with information derived from other sources such as TomTom. In the case of Flickr, OpenStreetMap fills a void in areas like Baghdad and Beijing where it can be harder to get access to map data.
While it’s nice to see OpenStreetMap’s crowdsourced model getting external validation through these applications, there’s a greater benefit. Having all these commercial services relying on OpenStreetMap data helps to place greater emphasis on the quality of data being added to the map. As long as this is balanced with the needs of other stakeholders such the American Red Cross and Liga Peatonal, OpenStreetMap will be the best geographic representation of the world around us.
What are we up to?
Filtering by benches detected in Aizuwakamatsu
On our side, we continue to work with the community and other partners to build the most comprehensive and diverse dataset of street-level imagery available. From this imagery we are able to detect temporal objects such as cars, and more permanent objects such as benches and crosswalks. The latter is useful for OpenStreetMap in two particular ways.
Firstly, being able to visualize detections on the map improves the map editing workflow, allowing the editor to focus on a particular map attribute and map it wherever Mapillary imagery is available. Secondly, and in a similar vein to Facebook’s work with satellite imagery, our detections can bring emphasis to map attributes that have had less attention up to this point. Detections such as the location of curb cuts and sidewalks are of particular importance to pedestrian routing, and our ability to identify them using machine learning is continually improving.
All of the AI work taking place compliments the role individuals have played since day one, using Mapillary to photograph and then map the places they care about.
Hot and sweaty after a quick photo walk around the castle grounds
State of the Map 2017 was well worth the journey. Thanks to both the local and international organisers who ensured the conference went as smoothly as it did, and to the delegates from every continent who shared their insights. Mapillary looks forward to seeing you again in Milan in 2018!