Source
Introducing OlmoEarth embeddings: Custom embedding exports from OlmoEarth Studio for downstream analysis
OlmoEarth Studio now supports the computation and export of embedding vectors from open source OlmoEarth foundation models for analyzing Earth-observation data.
Why it matters
This allows users to perform geospatial tasks like land-cover mapping or change detection more efficiently using far fewer labeled pixels than traditional methods.
The details
- Embeddings are exported as Cloud-Optimized GeoTIFFs stored as signed 8-bit integers.
- Supported imagery sources include Sentinel-1 RTC and Sentinel-2 L2A.
- Available encoder variants include Nano, Tiny, and Base models.
Show entities and relationshipsHide entities and relationships
In this article
Topics
Technologies
Products
Countries
Companies
Organizations
Key connections
Microsoft owns Microsoft Planetary Computer
Microsoft operates the Microsoft Planetary Computer platform.
European Space Agency owns Sentinel-2
The European Space Agency operates the Sentinel-2 satellite mission.
European Space Agency owns Sentinel-1
The European Space Agency operates the Sentinel-1 radar satellite mission.
European Space Agency owns ESA WorldCover 2021
The European Space Agency produces the ESA WorldCover 2021 dataset.
OlmoEarth Studio uses OlmoEarth
OlmoEarth Studio computes and exports embedding vectors using OlmoEarth foundation models.
OlmoEarth Studio uses Microsoft Planetary Computer
OlmoEarth Studio accesses Sentinel imagery via Microsoft Planetary Computer.
Show 13 more connectionsShow fewer connections
OlmoEarth Studio uses Sentinel-2
OlmoEarth Studio uses Sentinel-2 L2A imagery for computing embeddings.
OlmoEarth Studio uses Sentinel-1
OlmoEarth Studio supports Sentinel-1 RTC imagery as a data source.
OlmoEarth Studio uses Cloud-Optimized GeoTIFF
OlmoEarth Studio exports embedding rasters as Cloud-Optimized GeoTIFFs.
OlmoEarth Studio uses Supervised Fine-Tuning
OlmoEarth Studio supports supervised fine-tuning for custom model training.
OlmoEarth uses Sentinel-2
OlmoEarth foundation models are pretrained on Sentinel-2 satellite imagery.
OlmoEarth uses ESA WorldCover 2021
OlmoEarth embeddings are evaluated against ESA WorldCover 2021 reference labels.
OlmoEarth is built with Principal Component Analysis
Principal Component Analysis is used to reduce and visualize OlmoEarth embedding dimensions.
OlmoEarth is built with Logistic Regression
Logistic regression classifiers are trained on OlmoEarth embeddings for few-shot segmentation.
OlmoEarth is built with Cosine Similarity
Cosine similarity on OlmoEarth embeddings powers similarity search and change detection.
OlmoEarth is built with Linear Probing
Linear probing is used to evaluate OlmoEarth foundation model representations.
OlmoEarth Studio is related to United States
OlmoEarth Studio examples include similarity search in Merced and change detection in Butte County, California.
OlmoEarth Studio is related to Vietnam
OlmoEarth Studio embeddings are evaluated on mangrove segmentation in Ca Mau, Vietnam.
OlmoEarth Studio is related to Netherlands
OlmoEarth embeddings are visualized via PCA over agricultural polders in Flevoland, Netherlands.
Related events
OlmoEarth Studio Launches Custom Embedding Exports for Earth Observation
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.