A dataset of building instances of typical cities in China

The performance of building automatic extraction is driven by a large number of training samples. To enrich the publicly datasets of cities in China, we created a building instance dataset that is sourced from high-resolution remote sensing images and was manually and interactively annotated.

This dataset consists of 7,260 samples of regions including 63,886 building instances in the cities of Beijing, Shanghai, Shenzhen and Wuhan, China. The annotations of the dataset consist of MS COCO 2017 format files and the corresponding building mask binary maps. This dataset provides fundamental data for the research of building detection and extraction from high-resolution remote sensing images.