Lightning Talk: TerraKit: Standardising AI-Ready Geospatial Data... Rosie Lickorish & Romeo Kienzler

This talk introduces TerraKit, an open-source Python library designed to standardize the preparation of AI-ready geospatial datasets for the TorchGeo ecosystem. The speakers, Rosie Lickorish and Romeo Kienzler from IBM, discuss how TerraKit retrieves and processes geospatial data from various sources like EarthData and Planetary Computer, allowing users to define their specifications through an easy configuration file. This library not only facilitates raster and vector annotations but also supports spatial train/validation splits and exports datasets in formats such as TACO. By streamlining the transition from raw Earth observation data to AI-ready datasets, TerraKit accelerates the development of custom geospatial applications, making them more accessible to a broader community of users involved with TorchGeo and TerraTorch.