keras/dinov2_base

Model Overview

Vision Transformer (ViT) model trained using the DINOv2 method.

Reference

• Learning Robust Visual Features without Supervision
• Vision Transformers Need Registers

DINOV2 offers a powerful, generalist visual backbone learned entirely from unlabeled images as described in DINOv2: Learning Robust Visual Features without Supervision

Links

• [DINOv2 Quickstart Notebook] - coming soon
• [DINOv2 API Documentation] - coming soon
• [DINOv2 Beginner Guide] - coming soon
• KerasHub Model Publishing Guide

Installation

Keras and KerasHub can be installed with:

pip install -U -q keras-hub
pip install -U -q keras

Jax, TensorFlow, and Torch come preinstalled in Kaggle Notebooks. For instructions on installing them in another environment see the Keras Getting Started page.

Presets

The following model checkpoints are provided by the Keras team. Weights have been ported from: https://huggingface.co. Full code examples for each are available below.

Preset name	Parameters	Description
dinov2_small	22.58M	Vision Transformer (small-sized model) trained using DINOv2.
dinov2_base	87.63M	Vision Transformer (base-sized model) trained using DINOv2.
dinov2_large	305.77M	Vision Transformer (large-sized model) trained using DINOv2.
dinov2_giant	1.13B	Vision Transformer (giant-sized model) trained using DINOv2.
dinov2_with_registers_small	22.58M	Vision Transformer (small-sized model) trained using DINOv2, with registers.
dinov2_with_registers_base	87.63M	Vision Transformer (base-sized model) trained using DINOv2, with registers.
dinov2_with_registers_large	305.77M	Vision Transformer (large-sized model) trained using DINOv2, with registers.
dinov2_with_registers_giant	1.13B	Vision Transformer (giant-sized model) trained using DINOv2, with registers.