DINOv3
Extracts dense visual features from ordinary photographs, used as a frozen backbone for downstream tasks.
- Field
- Cross-domain
- Method
- Foundation model
- Data it takes
- imagery
- License
- LicenseRef-DINOv3 (not open source) — restricted use
DINOv3 is a ViT-L/16 backbone trained on web imagery. It produces 1024-dimensional per-patch tokens that a downstream head consumes — a linear probe, a segmentation head, a retrieval index.
This is the general-image counterpart to DINOv3-SAT. Use it for ground photographs, microscope images and laboratory imagery, and use DINOv3-SAT for aerial and satellite orthoimagery. The two share an architecture and differ in training data, and their preprocessing differs with it: DINOv3 expects ImageNet channel statistics, and supplying the satellite normalization instead degrades the features without raising an error.
The weights come under Meta’s own license rather than an open-source one. Publications reporting results must acknowledge DINO Materials.
Catalog entry last checked 2026-09-08. All models →
Run DINOv3 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