OlmoEarth Studio now allows users to compute and export embedding vectors from its open-source OlmoEarth foundation models for use in downstream tasks. These compact numerical representations are delivered as lightweight Cloud-Optimized GeoTIFFs (COGs) that can be shared and analyzed using standard geospatial tools.

  • Users configure exports via the Studio UI or API by selecting an area of interest, time span, encoder variant, resolution, and imagery sources.
  • Encoder variants include Nano (128-dim), Tiny (192-dim), and Base (768-dim), with vectors stored as signed 8-bit integers for efficient storage.
  • The embeddings support applications such as similarity search, few-shot segmentation, change detection, and unsupervised exploration without requiring task-specific training.
  • For higher performance needs, the platform also supports supervised fine-tuning (SFT) on top of these frozen features.

This feature provides a fast, cost-effective entry point for leveraging OlmoEarth by enabling users to generate custom representations that reflect specific conditions and time periods.