This repository contains the code supporting the DINOv2 base model for use with Autodistill.
DINOv2, developed by Meta Research, is a self-supervised training method for computer vision models. This library uses DINOv2 image embeddings with SVM to build a classification model.
Read the full Autodistill documentation.
Read the DINOv2 Autodistill documentation.
To use DINOv2 with autodistill, you need to install the following dependency:
pip3 install autodistill-dinov2from autodistill_dinov2 import DINOv2
target_model = DINOv2(None)
# train a model
# specify the directory where your annotations (in multiclass classification folder format)
# DINOv2 embeddings are saved in a file called "embeddings.json" the folder in which you are working
# with the structure {filename: embedding}
target_model.train("./context_images_labeled")
# get class list
# print(target_model.ontology.classes())
# run inference on the new model
pred = target_model.predict("./context_images_labeled/train/images/dog-7.jpg")
print(pred)The code in this repository is licensed under a CC Attribution-NonCommercial 4.0 International license.
We love your input! Please see the core Autodistill contributing guide to get started. Thank you 🙏 to all our contributors!