File size: 922 Bytes
ee1dfc7 c0ab1a1 ee1dfc7 f5d6e2c ee1dfc7 c0ab1a1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 | ---
title: DINOv2 Vision Demo
emoji: 🦕
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: "6.9.0"
python_version: "3.11"
app_file: app.py
pinned: false
license: apache-2.0
---
# DINOv2 Vision Demo
A lightweight demo of Meta's **DINOv2** self-supervised vision transformer, running entirely on CPU.
## Features
- **Image Classification** — Classify images against ImageNet-1k using `dinov2-small-imagenet1k-1-layer`.
- **Feature Visualization** — PCA projection of DINOv2 patch tokens rendered as an RGB map, revealing how the model perceives structure and semantics.
- **Image Similarity** — Compare two images via cosine similarity of their CLS token embeddings.
## Models
| Task | Model | Size |
|---|---|---|
| Classification | `facebook/dinov2-small-imagenet1k-1-layer` | ~86 MB |
| Features & Similarity | `facebook/dinov2-small` | ~86 MB |
All inference is performed on CPU — no GPU required.
|