Utonia

Creates embeddings from remote-sensing point clouds without requiring labelled training data.

Field
Cross-domain
Method
Foundation model
Data it takes
point-cloud
License
Apache-2.0 code, CC-BY-NC-4.0 weights — restricted use

Utonia continues the line that runs through Point Transformer V3, Sonata and Concerto, and it is the first of them pretrained across remote-sensing point clouds. Sonata’s pretraining is dominated by indoor scans, so Utonia starts closer to airborne and terrestrial survey data.

Like the others it is an encoder. The embeddings feed a classifier or segmentation head trained on a small labelled set.

Callers must pass enable_flash=False, because the image is built without flash-attention. The code is Apache-2.0 but the weights are non-commercial, and the weights are what binds a user.

Upstream project → · Paper →

Catalog entry last checked 2026-09-08. All models →

Run Utonia 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