CD-UGSD 1.0: A Remote-Sensing Dataset for Binary Urban Greenspace Extraction

CD-UGSD 1.0 is a remote-sensing dataset constructed for binary urban greenspace extraction in the built-up area of Chengdu, China. The dataset provides high-resolution PNG image tiles and corresponding pixel-level semantic segmentation masks for binary greenspace mapping. It is intended for academic research on urban greenspace identification, remote-sensing semantic segmentation, land-cover extraction, and method comparison under complex urban scenes. The dataset contains 60 image-mask pairs. Each image tile has a size of 3968 × 3968 pixels. The images are stored in PNG format, and the corresponding masks are also stored in PNG format. The image year is 2020, and the annotation year is 2023. The coordinate reference system is WGS 1984, and the imagery follows the projection exported from Google Earth Engine. The dataset follows a VOC-style semantic segmentation label convention. The released masks use 0 for background and 1 for greenspace. The value 255 is reserved for ignored or void pixels if present. Label definition:0: Background, including non-greenspace areas.1: Greenspace, including tree canopy, grassland, parks, and other urban green areas.255: Ignore / void, used only for ignored pixels if present. This release does not provide a fixed official train/validation/test split. Users may define their own split according to their experimental design. For reproducibility, users should report the number of images in each split, the random seed if random splitting is used, and the exact filename lists used for training, validation, and testing. The dataset is released for academic research only. Commercial use is prohibited. The dataset should not be interpreted as an official map product or an authoritative greenspace boundary product. Project repository: https://github.com/1kem1/Chengdu-Greenspace-Dataset