A unified multi-modal Earth Observation pre-training dataset combining Sentinel-2, Landsat 8/9, Copernicus DEM, and ESA WorldCover on a global 10 km grid. 250,000 tiles, TACO v3 format.
MajorTOM Elliot-Pretrain is the first expansion of Major TOM focused on fast pre-training of multi-modal AI models on Major TOM data.
This dataset is designed to grow incrementally -- we start with 250,000 monotemporal tiles and are going to add time-series data in the near future. The dataset follows the TACO v3 specification, a format for organizing AI-ready Earth Observation datasets.
Pick a tile index and visualize all four modalities:
A complete notebook with metadata queries, filtering, and a streaming PyTorch DataLoader with parallel fetching is available here:
CC-BY-SA-4.0
MajorTOM-Core has been made possible thanks to Asterisk Labs, the ELLIOT project (European Commission, Horizon Europe, Grant 101214398), and the Image and Signal Processing Group (ISP) at Universitat de Valencia.