PaperarXiv preprint · 2025
DINOv3
O. SiméoniH. V. VoM. SeitzerF. BaldassarreM. OquabC. JoseV. KhalidovM. SzafraniecS. YiM. RamamonjisoaF. MassaD. HazizaL. WehrstedtJ. WangT. DarcetT. MoutakanniL. SentanaC. RobertsA. VedaldiJ. TolanJ. BrandtC. CouprieJ. MairalH. JégouP. LabatutP. Bojanowski
1page built on it
0more pages cite it
1narrative
0registry papers cite it
In the Atlas
ModelPrimary source of
DINOv3
A 6.7B-parameter self-supervised ViT trained on 1.7B curated images, with gram anchoring to stop dense-feature degradation at scale — one frozen backbone whose patch features match or beat specialized fine-tuned pipelines on detection, segmentation, and depth.
Appears in 1 narrative
Lineage
Papers in the Atlas registry only · one dot per paperBuilt upon by · 0