cybergis/rs-embed

Provides unified access to 20+ remote sensing foundation models (SatMAE, Prithvi, DOFA, etc.) with automatic data fetching from Google Earth Engine, supporting both precomputed embeddings and on-the-fly inference across Sentinel-2, Sentinel-1, and multi-spectral sources. Built on PyTorch with TorchGeo integration, it abstracts spatial/temporal specification and model-specific preprocessing into a simple API that handles resolution conversion, band selection, and output pooling. Targets applications like land-use classification, yield prediction, and change detection by enabling batch inference across multiple regions and time periods.