Facebook unveils AI tool capable of mapping the world
What's the story
Facebook dominates as a social media giant, but its work in the technology industry goes way beyond that. The company has been exploring a range of technologies, mostly revolving around AI and machine learning. And now, as part of the same effort, it has unveiled AI-powered tools capable of mapping the whole world. Here's all about it.
Project
What exactly has Facebook done
Over the last two years, Facebook has been working on 'Map with AI' to develop machine learning-based tools for OpenStreetMap, the organization responsible for creating publicly-available, crowd-sourced maps. The members of OpenStreetMap community contribute towards improving the maps, and now, with Facebook's tools at their disposal, they can find previously unknown routes, places and do the same job easily.
Details
What these tools do?
While all these tools help in creating maps, the most essential one of the lot, RapidID, handles the task of identifying and adding roads into the maps. It uses satellite imagery provided by US space giant Maxar and predicts where roads might be present in a given area. The system marks new roads with a magenta overlay while existing ones are shown in white.
Information
Then, community members can add roads into Maps
Once a new road is marked with the magenta overlay, a member of the OpenStreetMap community can click on it to add it into the publicly-available map. The process is relatively easy.
Training
How the AI system was trained
In order to achieve mapping accuracy, the folks at Facebook trained an AI model on existing roads on Maxar's maps. Then, using that model, they developed a 34-layer deep neural network (DNN) capable of identifying roads at two square feet per pixel in a satellite image. It can give results with extremely high accuracy, but human judgment would still be needed for finalization.
Hope
Ultimately, this could help with disaster response and planning
Facebook has tested the tools to map 300,000 miles of unmapped roads in Thailand. They say "it can spot unpaved roads, as well as alleys and even pedestrian pathways, and distinguish them from visually similar riverbeds or walls." Now, as many parts of the world are difficult to map, this would help the community to find routes and roads in say a disaster.