taylor-geospatial/MINDSET

MINDSET MINDSET is the pretraining dataset for MIND, a coordinate-only location encoder distilled from static location encoder teachers and annual AlphaEarth Foundations (AEF) embeddings. We release the embeddings at the 12.1M training coordinates. The dataset contains 12,099,072 land coordinates in WGS84. Coordinates are dense around cities and not uniformly sampled over land. The files are in GeoParquet format and can be joined on point_id: file grain rows columns… See the full description on the dataset page: https://huggingface.co/datasets/taylor-geospatial/MINDSET.