Home Blogs Facebook opens up its AI-powered mapping service for OSM community

Facebook opens up its AI-powered mapping service for OSM community

Facebook OSM AI mapping
The road network around Mount Muria, Indonesia on Maxar satellite image.

Facebook has opened up its Map With AI service to the OSM community. This means anyone can have access to Facebook’s AI-powered mapping tools for a higher level of detail, quality and accuracy while creating maps on OSM. Initially, mappers will have access to its AI-generated road maps in Afghanistan, Bangladesh, Indonesia, Mexico, Nigeria, Tanzania and Uganda with more countries to be rolled out over time, Facebook said.

The move is in continuation of Facebook’s efforts to bring map the unmapped and ensure connectivity in the entire world. In May this year, Facebook released AI-powered population density maps to help humanitarian organizations respond to emergies and disasters.

“Over the past year we have been getting requests from many organizations asking for AI-generated roads and populations density maps, so we have been testing our tools with them to make sure it meets their needs,” Facebook said in a statement.

ALSO READ: Missing Maps Project uses tech to aid relief work in disaster-hit areas

What is Map With AI?

Map With AI includes an editor interface, RapiD, which allows mapping experts to easily review, verify, and adjust the map as needed. RapiD is an enhanced version of the popular OSM editing tool iD and is designed to make adding and editing roads quick and simple for anyone to use. It also has an extensive set of validation tools to help users catch and fix data issues in real time, thus improving the quality of the final submissions to OSM.

Facebook OSM AI mapping
The RapiD editor shows existing OSM roads (white) and newly detected ones (magenta) on top of Maxar satellite imagery. Image courtesy Facebook
Facebook OSM AI mapping
Left: results of the segmentation model per-pixel predictions; bright magenta means higher probability of the pixel belonging to a road. Right: Conflation of the vectorized roads data with the existing OSM roads (in white) on Maxar satellite images. Courtesy Facebook

The Facebook team has been working with local communities and partners to perfect the technology and in just 18 months was able to map missing roads in Thailand and more than 90% of missing roads in Indonesia. AI-powered mapping approach had never been deployed at such a scale before.

According to Xiaoming Gao, a Facebook research scientist who helped lead the project, mapping these vast unmapped areas in the traditional way without AI would have taken three to five years. “We were really excited about this achievement because it has proven Map With AI works at a large scale,” Gao says.

Now, the Map With AI team is collaborating with Humanitarian OpenStreetMap Team (HOT)  to add more features to RapiD such as integration it into a development branch of HOT Tasking Manager, which pairs volunteer mappers with specific areas to map, as an early experiment.

ALSO READ: Facebook launches maps for health organizations to respond to emergencies

Machine learning on satellite imagery

Facebook OSM AI mapping
Road extraction from a relatively well-mapped area in Kampala, Uganda. From left to right: Maxar satellite imagery, OSM (manually mapped), THA/IND/IDN trained model, Global OSM trained model. The model trained on DeepGlobe draws numerous nonexistent roads through the middle of houses, whereas the globally trained model performs well. Image courtesy Facebook

While developing the model, the Facebook team used Maxar satellite imagery to build the computer vision model for detecting roads. The roads were first traced by hand, and then used to teach the machines what was a road and what wasn’t. Mappers can then use the AI system’s predictions as the basis wherein the AI does majority of the work and mapper only filling in any gaps, double-checking accuracy and select appropriate road types.

Facebook OSM AI mapping
Visualization of the geographic distribution of training data for the OSM road segmentation model. Some areas are missing because satellite imagery was unavailable at the time of the experiments. Image courtesy Facebook

The 34-layer Deep Neural Network (DNN) model built by the Facebook AI team can recognize roads on satellite images around the globe with a resolution of roughly 2 square feet per pixel. This level of detail means it can spot unpaved roads, as well as alleys and even pedestrian pathways, and distinguish them from visually similar riverbeds or walls.

The Facebook-Maxar partnership goes way back in 2015 wherein the social media giant uses the latter’s Vivid imagery mosaic product for its highest resolution and most visually consistent global image mosaics available on the market.

One obvious reason for this partnership was the high accuracy of Maxar’s satellite imagery when it came to matching coordinates of the intersections in the imagery with as it on Earth. This provides a reliable layer with which to align OSM vector content (nodes, ways and the relations between them).

Now, the Map With AI service will have the latest of Maxar’s Premium and Standard Imagery layers on OpenStreetMap for everyone to use for mapping with the tool.

ALSO READ: Google Maps announces real-time disaster alerts and crisis navigation