Table 1 Existing typical large-scale building datasets.
Dataset | Source/Time Span | Coverage | Methods | Resolution | Type |
|---|---|---|---|---|---|
Microsoft BRA20 | Bing map; No time span | Not including China | DNN | Vector | Rooftop |
Google BRA23 | Google map; No time span | Africa / South Asia and Southeast Asia / Latin America | U-net | Vector | Rooftop |
CBRA21 | Sentinel 2; 2016–2021 | China | STSR-Seg | 2.5 m | Rooftop |
90_cities_BRA24 | Google Earth satellite; 2020 | 90 cities in China | Deeplab-V3 | Vector | Rooftop |
East Asian buildings22 | Google Earth satellite, GUB2018; 2022 | China, Japan, South Korea, North Korea and Mongolia | CLSM | Vector | Rooftop |
EUC41 | Landsat and Sentinel-1 SAR; 2015 | Europe, USA, China | Machine learning | 1 km2 | Height |
Wu et al., 202329 | Sentinel 1-2, PALSAR, LUOJIA1-01; 2020 | China | Machine learning | 10 m | Height |
Northern Hemisphere25 | Sentinel-1/2 images; Google Earth satellite; 2020 | China, the conterminous United States (CONUS), Europe | SRHS | 2.5 m | Height |
GABLE28 | Beijing-3 satellite imagery, WSF2019; 2023 | China | RPN | Vector | Rooftop and height |
3D-GloBFP26 | Microsoft BRA; 2020 | Global | Machine learning | Vector | Height |
Zheng et al., 202424 | East Asian buildings, Baidu, OSM, Gaode; No time span | Three major urban agglomerations in China | Machine learning | Vector | Function |
CMAB(Ours) | Google Earth satellite, Spatial Cites; 2021–2024 | China | OCRNet, XGBoost, Yolov8, LMMs | Vector | Rooftop, height, structure, function, style, age and quality |
