Meta’s self-supervised vision transformers (DINO, DINOv2, DINOv3). Their dense features are used for matching, place recognition and segmentation without labels.
Mentioned in
8 papers:
- Hickory (2026)
- TrackGraph (2026)
- Hydra++ (2026)
- Mono-Hydra++ (2026)
- Depth Anything 3 (2025)
- FunGraph (2025)
- Opti-acoustic semantic SLAM (2024)
- OpenGraph (2024)
Notes: