DINOv2 is a Meta vision model from 2023 that learns general-purpose image features from pictures alone, with no labels or captions.
In plain words
DINOv2 uses self-supervised learning: it learns from the images themselves, with no labels or captions attached. Its authors showed that doing this on a large, carefully curated image collection produces features good enough to use straight away. You can attach a simple linear classifier on top for many tasks, across different kinds of images, without fine-tuning the model itself.
A quick example
Imagine a botanist with a few hundred labelled leaf photos. They run each photo through DINOv2 to get its features, then train only a small linear classifier on those features to tell the species apart.
Where you'll meet it
- A ready-made feature extractor for many vision tasks.
- Computer vision research.
Related
Sources
- PaperDINOv2: Learning Robust Visual Features without Supervision, Oquab et al. (arXiv) · read 29 Sept 2026
- Docsdinov2, Meta (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.