RapiD mapping Saikat Basu, Derrick Bonafilia, James Gill, Danil Kirsanov, David Yang/Facebook AI Some of us take for granted our road maps, not necessarily acknowledging just how much work goes into them, and how quickly they can become outdated due to natural disasters, time, or simply new infrastructure.
Facebook AI has been working well and truly hard to create computer and AI systems to facilitate the road mapping process around the world.
Most maps are created for highly developed areas but don’t take into account the majority of the world – which relies on dirt or gravel roads, or unpaved paths. — #AI (@AI__TECH) July 23, 2019 Google Maps and Apple have certainly been trying and perhaps doing their best at providing road maps, but their focus has been mostly on navigating big cities, allowing drivers, cyclists, public transport commuters and walkers to get to well-known businesses and addresses.
Now Facebook is stepping in to help facilitate the bigger, less traveled upon, picture, working closely with OpenStreetMap.
Facebook engaging with the public for help
It doesn’t simply take a team of researchers in a back room to update a street map. What Facebook is looking for is people on the ground, willing and able to assist it with improving its modern mapping services. RapiD mapping experience. Source: Facebook It’s working on the project closely with OpenStreetMap (OSM) and helping validate their roads.
A perfect example of this has been Facebook’s mapping of over 300,000 miles of roads across the entirety of Thailand . This, in turn, created RapiD, a machine-learning enhanced labeling tool that accelerates the process of putting down computer-readable roads onto satellite images.
RapiD is an open-source extension of the web-based iD map editor, and it allows human reviewers to work on the maps. This helps with mapping out accurate road systems, and there are safety checks in place in order to ensure top quality results.
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).(Satellite images provided by Maxar.) It’s very important to decipher exactly which of the highlighted sections are indeed roads. The AI systems assist to validate these roads, with mostly images taken from satellites and ensure they’re accurately positioned, as many can be mistaken for dry riverbeds, for example.
Anyone can help map the world, by simply joining the OSM troops .