Potsdam Dataset for Semantic Segmentation v1.0

The Potsdam dataset is a high-resolution remote sensing benchmark for 2D semantic labeling, provided by ISPRS WG III/4. It contains 38 orthophoto tiles (6000×6000 pixels each) with a spatial resolution of 5 cm/pixel, covering the city of Potsdam, Germany. File Structure: Ortho_IRRG: Infrared-red-green false-color images (TIFF, 3 bands) DSM_normalisation: Normalized DSM ( normalized to ground level) Labels_for_participants: Semantic segmentation labels (TIFF, 6 classes) Technical Specifications: Spatial Resolution: 5 cm/pixel Coordinate System: UTM 33N (EPSG:32633) Image Dimensions: 6000 × 6000 pixels per tile Label Classes: 6 classes 0: Impervious surfaces 1: Building 2: Low vegetation 3: Tree 4: Car 5: Background Usage Notes: Image and label files share the same naming convention Recommended libraries: GDAL, rasterio (Python), OpenCV