From the model card
Fields below are copied from the tags and counters on the HuggingFace repository facebook/dinov2-small at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- Pipeline tag
- image-feature-extraction
- Library
- Transformers
- Framework tags
- PyTorch
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Papers cited
- arXiv:2304.07193
- Downloads (HF counter at last fetch)
- 4,372,990
- Likes (HF counter at last fetch)
- 72
- Model card
- https://huggingface.co/facebook/dinov2-small
Use cases
- Lightweight visual feature extraction for downstream classification
- Edge vision inference where ViT-B is too large
- Fast ablation studies across the DINOv2 scale ladder
- Image retrieval in applications with tight latency budgets
Pros
- 21M parameters enable fast inference and low memory use
- Apache-2.0 licensed
- Part of a consistent scale family — easy to upgrade to base/large
- Self-supervised training means no label dependency
Cons
- Significantly weaker features than ViT-B on most tasks
- Not suitable for dense prediction tasks that benefit from larger capacity
- Small model saturates quickly when applied to fine-grained visual domains
- Diminishing returns on compression vs using quantized ViT-B
Tags
transformerspytorchsafetensorsdinov2image-feature-extractiondinovisionarxiv:2304.07193license:apache-2.0endpoints_compatibleregion:us