UBC-CN: Fine-Grained Building Extraction Dataset For Chinese Regions

In this paper, we propose a new large scale dataset dedicated to fine-grained building extraction in Chinese regions. To ensure the representativeness of the dataset samples, various factors are taken into consideration, encompassing geographical distribution, appearance variety, rural areas and spatial layout. Consequently, the dataset is composed of 131 K building instances, covering an expansive area of 784.3 square kilometers across 34 cities. Each building is represented with a polygon and a corresponding rooftop type. The rooftops are categorized into 12 fine-grained types. Moreover, we conduct experiments using four classical methods on this dataset to establish baselines for future studies. This dataset will promote the refined development of the city management and planning by providing essential resources for developing the state-of-the-art methods to accurately identify building instances on a city or national scale.