Replicable AI for Microplanning - A DevGlobal Initiative, in partnership with the World Health Organization.
The ramp dataset series is a collection of building footprint training data across diverse geographies. The series include 22 individual training datasets covering specific regions or cities. Each ramp dataset includes high-resolution imagery in a TIFF format and corresponding building footprint vector labels in a GeoJSON format. The imagery is divided into 256 x 256 pixel tile label pairs, which have been annotated by an individual and reviewed by another for quality control.
The datasets are available under a CC BY-N.C. 4.0 license, and DevGlobal plans to release more imagery chips and labels in the future.