Deep learning and weakly supervised training to assist Facebook AI in mapping roads of developing countries.

Accessing satellite images and creating accurate maps is a manually time-consuming process. Most of the regions of the developing world has remained unmapped. This issue has been taken up by Facebook with its AI technology towards a resolution which appears to be upcoming in recent years. 

Facebook Artificial Intelligence in a blog post revealed its new method which has been developed by its AI researchers and engineers to predict road maps from the available commercial high-resolution satellite imagery. The Maps with AI includes an editor interface, RapiD, that allows the experts to review, verify and adjust the necessary maps. Facebook states that they had used the system previously to map the roads of Thailand which were more than 300,000 miles in the stretch in OpenStreetMap which is a community-based effort intended to make clear, free and editable maps of the world and spots. 

A team of mapping experts and AI researchers took just 18 months to completely map the streets of Thailand. This was less than half of the expected interval of practice. Facebook acknowledged that Accuracy in the mapping will help it serve a larger audience with services like Facebook Marketplace and Facebook Local. Moreover, with the AI population density maps project, these maps will be made public for disaster response, urban planning, developmental projects, and many more utilities. 

Facebook addresses the fact that it has implemented a number of new techniques for more efficiency of the service. The company reminds us of CVPR 2018 when they organized the DeepGlobe Satellite Challenge to collect image analysis on satellite generated imageries. They then evaluated the computer vision and machine learning solutions, developing new learning techniques and architectures designed to suit the problem space of remote sensing.

 “Our efforts are focused on building RapiD, an open-source extension of the widely used web-based iD map editor. Additionally, we built a system that combines the model’s results with data already available in OSM. This process, called conflation, both advises on how to join new roads with existing data and prevents overwriting existing road data with suggested roads. It is our hope that RapiD will allow people in the mapping community to improve and leverage these tools for their own use cases.”

Facebook’s Official Blogpost

With good tooling, the team looks forward to empowering the mappers and effectively reduce the tedious and time-consuming sector of drawing the roads with satellite images and increased road shape accuracy and provide options for identifying suggested roads. 

Facebook has taken special care of the fact that their mapping should not limit the capabilities and judgment of professional mappers due to AI implementations. RapiD will be improved in the basis of feedback from the mappers do make the process glitch-free and smoother than ever. 


Mary Deshazo

Mary is a food and mobile tech industry enthusiast. She sleeps an eye open looking for industry updates and spends weekends fishing with her husband.

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button
Deep learning and weakly supervised training to assist Facebook AI in mapping roads of developing countries., Tech chums