Map With AI announces a new partnership with Microsoft, expanded coverage, and new tools

Today, we are sharing new global coverage for RapiD in nearly every country in the world, a new partnership with Microsoft Buildings, and collaboration on a new Map With AI plugin to support Java OSM Editor (JOSM).

Expanding RapiD globally

When it first launched, Map With AI included roads available for mapping in nine countries. In the months since then, the team has received more than a hundred requests from both organizations and individuals to add coverage for additional countries. In response, we have continued rolling out additional coverage in waves. Our initial plan was to roll out three countries per month as we assessed our pipeline’s scalability. With the enthusiastic response to Map With AI, we scaled up more quickly to produce closer to 90 countries per month, which allowed us to crank out nearly the entire world by the end of 2019.

Response from the OSM community has been very enthusiastic. Mappers from 137 countries have used the tool, contributing over 100,000 changesets to the map. More than 200,000 kilometers of new roads have been added from more than 1,000,000 new features.

We also received multiple requests to make Map With AI road data available for direct download. A major use case was humanitarian mapping, such as Crowd2Map Tanzania, using our data to plan road mapping projects. These country-scale road exports can identify unmapped areas to better guide organized mapping efforts and increase mapper efficiency, as areas without any predicted roads can be excluded from projects. Mapping enthusiasts have also taken note of Map With AI. For many countries, Map With AI combined with OSM provides the most complete road data publicly available. “Map With AI data provides so much opportunity for the mapping community, as well as the people who live in these countries,” says Justin Meyers, a GIS data analyst at AccuWeather. “Many nations have no formal mapping agency, and open data is years away. Seeing cities come to life from new AI traces allows previously unmapped communities to now be on the map.”

Adding JOSM support

Support for JOSM was one of the update requests we heard frequently after launching RapiD. JOSM is a powerful OSM editor favored by experienced mappers for its flexibility and speed, but it can be intimidating for new users to learn. Happily, Kaart developer Taylor Smock took the initiative, started building a Map With AI plugin, and then presented an early prototype to Facebook last September. Thanks to a collaboration with Smock, we are now able to release the completed plugin for public use.

With JOSM support, AI-assisted mapping has now been made available to even more mappers, including power users who want to access our map data but prefer not to change tools. Since RapiD requires a constant internet connection, it is not ideal for parts of the world where internet connections are slow or intermittent. JOSM, by contrast, downloads all the map data initially and allows mappers to work offline, which makes it far more convenient in areas with poor connections.

Since JOSM is more of a power tool, the plugin offers increased flexibility, like the ability to add multiple features at once and automatic retagging of AI roads. Combined with the functionality already built into JOSM, this means people can add roads and buildings faster than ever before. Since its launch in January, the plugin has been used in thousands of changesets, with usage rapidly rising.

Integration with Microsoft 

While our researchers have been hard at work perfecting AI-generated roads for the past few years, Microsoft has been busy developing their own map data for AI-generated buildings. They released this data set for the United States, Canada, Tanzania, and Uganda, but no tooling was available to make it easy to import into OSM. Coverage for buildings on OSM is very uneven, and coverage is highly lacking, even in developed countries where road coverage tends to be nearly complete. To help improve this situation, the goal was to make these AI buildings usable in OpenStreetMap.

We were interested in utilizing these buildings within Map With AI, so as an experiment, we integrated the Microsoft AI Buildings data set into RapiD. The results were very promising, as the buildings worked well with the same workflow we had created for roads. We reached out to Microsoft and presented our experiment. The team at Microsoft was happy to have their work made available in RapiD, and worked with Facebook to release the new version. 

Read more about the technology behind the integration with Microsoft.

The Map With AI team first demonstrated RapiD with Microsoft Building support at State of the Map US, an annual mapping conference, and the response was immediate and enthusiastic. People found that it made building mapping significantly faster and easier.

Considering future integrations

Beyond AI-based data sets, one of the biggest challenges for OSM is importing even readily available authoritative data sets. Many government agencies have map data available for free, but the process of creating an import is too onerous for many users, as the data must be converted to OSM tagging standards and conflating away duplicate data can be difficult and error-prone.

Today, more than 1,200 people and organizations are using the RapiD editor. We are excited to share these new developments to make it simpler and more efficient to produce accurate maps in a timely manner. Looking ahead, we intend to continue to expand coverage and to release new tools to support AI-powered mapping. As we continue to collaborate and to simplify the interface, our hope is that RapiD can become a tool that’s simple enough for anyone to import and verify new data sets and to make use of these powerful tools.

Visit Map with AI to take advantage of these new tools and features.

We'd like to thank Saikat Basu, Brian Bates, Anil Batra, Charmaine Bonifacio, Benjamin Clark, Xiaoming Gao, Nicholas Holroyd, Christopher Klaiber, Zack LaVergne, Yunzhi Lin, Peter Ouyang, Taylor Smock, Zvone Sparovec, and David Yang for their contributions to this project.