According to Bing Maps, the global building footprints around the world have been detected in 776,712,641 building footprint polygon geometries between 2014 and 2021 including Maxar and Airbus imagery. The dataset in line delimited GeoJSON format is freely available for download and uses under Open Data Commons Open Database License (ODbl) by Microsoft and coordinate reference system as EPSG: 4326. This released underlying data from Microsoft has a continued interest in supporting a thriving OpenStreetMap ecosystem and creating the data in two stages: 1/ Semantic segmentation-Recognizing building pixels on an aerial image under deep neural networks (DNNs) and 2/ Polygonization- Converting building pixel detections into polygons. With evaluation set metrics, buildings are computed on a set of building polygon labels for each region and matched metrics in the following precision: Africa 94.4%, Caribbean 92.2%, Central Asia 97.17%, Europe 94.3%, Middle East 95.7%, South America 95.4%, and South Asia 94.8%. However, the recall evaluation, there are Africa 70.9%, Caribbean 76.8%, Central Asia 79.47%, Europe 85.9%, Middle East 85.4%, South America 78%, and South Asia 76.7%. The source also states with measuring the quality of matched building polygons in intersection over Union and dominant angle rotation error.
In Cambodia, dataset count in building is with 3,874,425 and available to download via https://minedbuildings.blob.core.windows.net/global-buildings/2022-05-02/Cambodia.zip.
