paper: DINOv2: Learning Robust Visual Features without Supervision https://github.com/facebookresearch/dinov2