DINOv2: Learning Robust Visual Features without Supervision https://github.com/facebookresearch/dinov2 The weights can be extracted from pre-trained Python models using python src/vision/export_dinov2.py.
python src/vision/export_dinov2.py