CrossEarth

Segments satellite imagery in regions and conditions it was not trained on.

Field
Remote sensing
Method
Foundation model
Data it takes
imagery
License
MIT

Semantic segmentation models typically degrade when applied outside the region, resolution or climate they were trained on. CrossEarth targets that problem using a frozen DINOv2 backbone with augmentation designed to survive domain shift, and ships with a segmentation head.

It suits cases where labels exist for one area and the same classification is needed somewhere different.

Upstream project →

Catalog entry last checked 2026-08-17. All models →

Run CrossEarth on your data

The lab runs this on its own compute, in a container, with the inputs and parameters recorded alongside the result. Initial scoping conversations are free.

Contact the Lab