Satlas Super Resolution

Satlas Super Resolution is a component of the broader Satlas initiative, which aims to provide open, AI-generated geospatial data that is highly accurate, globally available, and updated on a frequent (monthly) basis. A key application within Satlas is the global generation of Super-Resolution imagery for 2023, enhancing the detail and clarity of satellite observations.

This GitHub repository contains the essential training and inference code necessary for generating the AI-powered Super-Resolution data, which is openly accessible via the main Satlas platform. It also includes the data and model weights corresponding to the research paper, "Zooming Out on Zooming In: Advancing Super-Resolution for Remote Sensing." The repository's main branch showcases core features, while an experiments branch provides configuration files for paper-related experiments.

The project facilitates installation via conda and pip, outlining dependencies for setting up the environment. It also provides access to two significant training datasets: an urban set with approximately 1.1 million Sentinel-2 image pairs from U.S. cities, and a comprehensive full set containing about 44 million pairs derived from NAIP imagery collected between 2019-2020, both available for download.