From the model card
Fields below are copied from the tags and counters on the HuggingFace repository timm/vit_small_patch14_dinov2.lvd142m 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)
- timm
- Pipeline tag
- image-feature-extraction
- Library
- timm, 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, arXiv:2010.11929
- Downloads (HF counter at last fetch)
- 1,135,437
- Likes (HF counter at last fetch)
- 8
- Model card
- https://huggingface.co/timm/vit_small_patch14_dinov2.lvd142m
Use cases
- Zero-shot visual feature extraction for downstream classifiers
- Dense feature maps for monocular depth estimation
- Image retrieval in visual search systems
- Semantic segmentation backbone without fine-tuning overhead
- Few-shot image classification via nearest-neighbour matching
Pros
- DINOv2 features are among the strongest self-supervised representations for vision tasks
- patch14 stride produces higher-resolution feature maps than patch16
- Apache 2.0 license; timm framework integration is seamless
- Small variant runs on modest GPUs with low latency
Cons
- ViT-Small has less capacity than ViT-Base/Large DINOv2 variants for complex scenes
- patch14 is non-standard; incompatible with most pre-built ViT pipelines expecting patch16
- Self-supervised features sometimes underperform supervised ones on narrow domains
- timm dependency adds a secondary version pinning requirement
Tags
timmpytorchsafetensorsimage-feature-extractiontransformersarxiv:2304.07193arxiv:2010.11929license:apache-2.0region:us