UrbanVGGT: Scalable Sidewalk Width Estimation from Street View Images

The exterior surfaces of buildings shape how cities use energy, store heat, and respond to disasters. However, this information is missing from most urban building datasets. BMAT fills this gap by mapping exterior wall materials for 22 million buildings in 73 cities from 147 million street-view images collected between 2007 and 2025. The open dataset, validated with 91 % accuracy, can support research on urban climate, energy use, planning, and disaster resilience.

China’s cities have transformed dramatically over the past 30 years. Using multi-source satellite data and machine learning, this study mapped annual building heights at 30-meter detail from 1990 to 2019, revealing both horizontal and vertical urban growth. The open dataset offers new insights into how Chinese cities expand and renew, supporting research and planning for urban development.