GeoDepth: a high resolution dataset for monocular depth estimation and camera self-calibration

GeoDepth is a high-resolution photogrammetric dataset for monocular depth estimation and camera self-calibration, developed using geological surface reconstructions. It consists of 8,272 high-resolution RGB images captured across diverse geological locations in Europe (Italy, Malta, Croatia) and the Middle East (Oman, Qatar), paired with precise 64-bit depth maps. Unlike existing RGB-D datasets that typically use limited bit-length depth images, GeoDepth retains the native resolution of various camera platforms (drones, DSLR/mirrorless cameras, smartphones) and provides complete camera calibration parameters essential for accurate 3D projection. The dataset covers geologically and morphologically diverse natural and manmade exposure surfaces, including quarries, cliffs, faults, and various rock formations. Each paired RGB-D image is supplemented with intrinsic camera calibrations required for coordinate transformation, making GeoDepth particularly valuable for high-precision depth extraction tasks in geospatial applications, geological analysis, and training monocular depth estimation models.