Urban Garden Ground‑Cover Dataset

Urban garden ground‑cover dataset comprises 10 high‑resolution orthomosaic images from two Munich community gardens, split into 6 train, 2 validation, and 2 test subsets. RGB images: 3‑band GeoTIFFs (EPSG:25832), 0.01 m ground sampling distance, 18.9–146.4 Mpx Ground‑truth masks: single‑band GeoTIFFs with eight semantic classes (grass, herb, litter, soil, stone, straw, wood, woodchip) To support deep learning, we provide an overlapping 512 × 512 px patch set (stride 256 px) for both images and masks, organized by the same train/val/test split. A comprehensive dataset_metadata.csv records each file’s dimensions, pixel count, spatial resolution, CRS, plus Agisoft processing details and flight properties.