Grounding DINO

Detects objects named in a free-text prompt, with no fixed class list.

Field
Cross-domain
Method
Foundation model
Data it takes
imagery
License
Apache-2.0

Grounding DINO takes a phrase and returns boxes for the objects it names. A new target is a new prompt, with no annotation and no training step in between.

It returns boxes only. Pairing it with SAM 2 converts those boxes into masks.

Detection quality follows how closely the prompt matches everyday phrasing. Specialist vocabulary works less reliably than common nouns, and a prompt often needs rewording before it finds what a specialist means by it.

Upstream project → · Paper →

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

Run Grounding DINO 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