chrieke/awesome-satellite-imagery-datasets

Organized as a curated taxonomy by task type (instance segmentation, object detection, semantic segmentation, scene classification), the collection spans multi-modal satellite sources—optical (Sentinel-2, Worldview-3), SAR (Capella, Sentinel-1), and temporal timeseries—with datasets ranging from synthetic benchmarks to large-scale production annotations (Microsoft's 125M building footprints). Includes associated competition challenges, baseline models, and papers linking each dataset to specific deep learning applications like change detection, damage assessment, and agricultural monitoring.

3,870 stars. No commits in the last 6 months.