AI Tools.

Search

image feature extraction by facebook

dinov2-base

DINOv2 ViT-B is Meta's self-supervised vision transformer trained on 142M curated images using a combination of DINO and iBOT objectives. It produces strong visual features for dense prediction tasks without any labels during pretraining.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository facebook/dinov2-base 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)
facebook
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)
2,843,962
Likes (HF counter at last fetch)
195
Model card
https://huggingface.co/facebook/dinov2-base

Use cases

  • Extracting general-purpose visual features for downstream tasks
  • Zero-shot image retrieval using embedding similarity
  • Semantic segmentation backbone fine-tuning
  • Few-shot classification with frozen features

Pros

  • Self-supervised — no manual labels required for pretraining
  • Features transfer well to segmentation, depth, and retrieval
  • Apache-2.0 licensed
  • Available in small/base/large/giant variants for different resource budgets

Cons

  • Requires fine-tuning a head for classification — not plug-and-play for prediction
  • Larger ViT variants (large/giant) provide substantially better features
  • No text alignment — can't do zero-shot classification like CLIP
  • Pretraining data curation pipeline is not fully reproducible

Tags

transformerspytorchsafetensorsdinov2image-feature-extractiondinovisionarxiv:2304.07193license:apache-2.0region:us