“Cataloging SpaceNet’s datasets help to expand Radiant MLHub’s global map of geospatial training datasets, which is essential to identify under-represented geographical areas”
Today, Radiant Earth Foundation announced the registration of a Spatio-Temporal Asset Catalog (STAC)-compliant version of SpaceNet’s high-quality geospatial labeled datasets for roads and buildings on Radiant MLHub. Radiant MLHub is the world’s first cloud-based open library dedicated to Earth observation training data for machine learning algorithms. The updated dataset catalog is also available through SpaceNet’s data registry.
Founded in 2016 to accelerate open source geospatial machine learning, SpaceNet is a nonprofit organization that runs data challenges and releases the training datasets, baseline algorithms, winning algorithms, and detailed evaluations under an open source license. They have organized six data challenges to date, each focusing on a different problem that applies machine learning to satellite imagery to solve complex mapping problems. The SpaceNet 7 Challenge is scheduled to launch early this Fall.
- Challenge 1 focused on extracting building features in WorldView-3 high-resolution satellite imagery in Rio de Janeiro. This dataset contains more than 382,000 building footprints.
- Challenge 2 focused on detecting building footprints in WorldView-3 high-resolution satellite imagery across four cities: Las Vegas, Paris, Shanghai, and Khartoum. There are more than 1.6 million building footprints across all four cities.
- Challenge 3 focused on detecting roads and their type in WorldView-3 high-resolution satellite imagery in Las Vegas, Paris, Shanghai, and Khartoum. Over 8,000 Km of roads with eight types are labeled across these four cities.
- Challenge 4 focused on extracting building footprints over 665 km2 of Atlanta, Georgia, in the US in WorldView-2 high resolution off nadir satellite imagery. The dataset contains more than 120,000 building footprints.
- Challenge 5 focused on road network extraction and route travel time estimation from high resolution satellite imagery in Moscow, Mumbai, San Juan, and a Mystery City. A total of over 8,100 Km of road labels is provided in these cities.
- Challenge 6 focused on detecting building footprints over an area of 120 km2 using multi-sensor high resolution Synthetic Aperture Radar (SAR) and electro-optical (EO) imagery. The dataset contains ~48,000 building footprints in Rotterdam, The Netherlands.
The training data for each challenge with the updated STAC catalogs are registered on Radiant MLHub as well as SpaceNet’s Open Data registry, providing easy access for data scientists working on urban planning and rural development problems. Cataloging SpaceNet’s datasets help to expand Radiant MLHub’s global map of geospatial training datasets, which is essential to identify under-represented geographical areas; it also helps automate the process of geospatial training data discovery which is a fundamental data management problem. Finally, it increases benchmarks for evaluating the accuracy of geospatial machine learning models.
“We are very excited to have SpaceNet’s training datasets from all six data challenges registered on Radiant MLHub,” said Radiant Earth Foundation Chief Data Scientist Hamed Alemohammad. “Radiant MLHub is a community resource for anyone working with Earth observations to discover machine learning ready data, publish their training datasets for added visibility, or benchmark algorithms against existing data. Partnering with leading organizations like SpaceNet helps us grow Radiant MLHub’s library.”
“Updating the SpaceNet data catalogs with the STAC standard is an important step to making current and future datasets easily accessible for practitioners,” said Ryan Lewis, General Manager of SpaceNet LLC. “We are grateful to the entire Radiant Earth team for their help in making this possible as well as making our data available on Radiant MLHub.”