kornia/dinov2

Pretrained DINOv2 ViT-L/14 backbone weights, used by kornia's DeDoDe descriptor encoder (kornia.feature.dedode.encoder.FrozenDINOv2, behind the G descriptors of kornia.feature.DeDoDe).

DINOv2 is a family of self-supervised vision transformers that give general-purpose visual features. This is the distilled ViT-L/14 backbone without heads.

Original repo: facebookresearch/dinov2

Weights

File	Model
dinov2_vitl14_pretrain.safetensors	ViT-L/14 distilled, backbone only

Provenance

dinov2_vitl14_pretrain.safetensors holds the tensors of the upstream dinov2_vitl14_pretrain.pth (sha256 d5383ea8f4877b2472eb973e0fd72d557c7da5d3611bd527ceeb1d7162cbf428), unchanged: the same keys, dtypes, shapes and bits. The kornia maintainers converted it (converter revision 0c3db846b2) and checked that the backbone loaded from either file gives bitwise-identical features. The sha256 of the converted file is a8a908b7571371a71f160ae436d7eb7b71689d9e6202c6751e533a256e420e81.

License

Apache-2.0. The upstream README states that "DINOv2 code and model weights are released under the Apache License 2.0". See LICENSE, copied from facebookresearch/dinov2. That README's noncommercial licence covers the separate XRay-DINO model, not these weights.

Citation

@misc{oquab2023dinov2,
    title   = {DINOv2: Learning Robust Visual Features without Supervision},
    author  = {Oquab, Maxime and Darcet, Timothée and Moutakanni, Theo and Vo, Huy V. and Szafraniec, Marc and
               Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin
               and Howes, Russell and Huang, Po-Yao and Xu, Hu and Sharma, Vasu and Li, Shang-Wen and Galuba,
               Wojciech and Rabbat, Mike and Assran, Mido and Ballas, Nicolas and Synnaeve, Gabriel and Misra, Ishan
               and Jegou, Herve and Mairal, Julien and Labatut, Patrick and Joulin, Armand and Bojanowski, Piotr},
    journal = {arXiv:2304.07193},
    year    = {2023}
}