DINO
- Self-Supervised Vision Transformers with DINO
- DINOv2: Learning Robust Visual Features without Supervision
- 2023-04
- DINO v2
- DINOv2: Learning Robust Visual Features without Supervision https://arxiv.org/abs/2304.07193
- Vision Transformers Need Registers https://arxiv.org/abs/2309.16588
- DINO v3
- 2025-08
- facebookresearch/dinov3
- https://arxiv.org/abs/2508.10104
- 在 web 数据上预训练(LVD-1689M 数据集):10 个模型
- 1 个 ViT-7B(从头训练)
- 5 个 ViT-S/S+/B/L/H+(蒸馏自 ViT-7B)
- 4 个 ConvNeXt-{T/S/B/L}(蒸馏自 ViT-7B)
- 在卫星数据上预训练(SAT-493M 数据集):2 个模型
- 1 个 ViT-7B(从头训练)
- 1 个 ViT-L(蒸馏自 ViT-7B)
- https://huggingface.co/facebook/dinov3-vit7b16-pretrain-lvd1689m