Sakhanda Wire
NVDA $153.77 -1.52% MSFT $506.06 -2.30% GOOGL $280.21 -1.24% META $622.96 -0.55% AMZN $257.51 -0.09%
← Back to the news

OlmoEarth Studio adds custom Earth-observation embedding exports

OlmoEarth Studio now allows users to compute and export embedding vectors generated by OlmoEarth foundation models. The compact numerical representations can support similarity search, land-cover segmentation, change detection and unsupervised analysis. The models’ source code and weights are publicly available alongside the research paper, allowing the generation process to be inspected.

Users can select an area, time span, encoder, spatial resolution and imagery sources through the Studio interface or API. Available encoders are Nano, with 128 dimensions and 1.4M parameters; Tiny, with 192 dimensions and 6.2M parameters; and Base, with 768 dimensions and 89M parameters. Resolutions range from 10 to 80 meters per pixel, and sources include Sentinel-2 L2A and Sentinel-1 RTC.

Results are delivered as Cloud-Optimized GeoTIFFs, with one band for each embedding dimension and signed 8-bit integer values. The article reports a weighted F1 score of 0.84 for a land-cover map created from 60 labeled pixels in Ca Mau, Vietnam. Studio also supports supervised fine-tuning for applications needing higher performance, while output quality can be affected by cloud cover, atmospheric artifacts and missing observations.

Written by Sakhanda Wire AI from reporting by Hugging Face. Facts, quotes and full context are in the original article below.

Originally published by Hugging Face on

Read the original on Hugging Face ↗

The reporting and the image belong to Hugging Face and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news