Microsoft Global ML Building Footprints
Microsoft (Bing Maps)
About 1.4 billion building outlines worldwide, traced by a neural network from aerial and satellite imagery.
Microsoft ran deep learning models over Bing Maps imagery from 2014 onward to detect building footprints and released the polygons openly, many with estimated heights. Files are partitioned by country and map tile. Humanitarian mappers, insurers, and researchers use them to fill gaps where official building data does not exist.
Access
- Cost
- FreeFree downloads from Azure storage via GitHub index; no account needed.
- License
- Open licenseCommunity Data License Agreement Permissive 2.0
- Sign-up
- No sign-up
- How you get it
- Bulk downloadGeoJSON (line-delimited)
What's in it
- One row is
- One polygon per detected building
- Covers
- Most of the world (225 regions)
- Years
- Imagery 2014-2025
- Updated
- Irregular
- Size
- About 1.4 billion buildings (the repository also cites 999 million in places)
Quality
Massive open coverage, but machine-detected outlines vary in accuracy and dates and include false positives.
- Imagery vintage varies by tile
- False positives and merged buildings in dense areas
- Height estimates missing for many buildings
Connects to
- Overture Maps: Overture's buildings theme merges these with OSM and Google footprints
- WorldPop Gridded Population: Building counts inform population distribution
Questions it could answer
- How many buildings sit in flood zones in a given country?
- Where are towns growing fastest in sub-Saharan Africa?