ProSMin: Probabilistic Self-supervised Learning via Scoring Rules Minimization

This repository is for "Probabilistic Self-supervised Learning via Scoring Rules Minimization" paper.

Table of contents

• Installation
• Dataset
• Usage/Examples
• Acknowledgements
• License

Dependency

Installation

Install requirments:

pip install -r requirements.txt

Dataset

Please download Imagenet.

Usage/Examples

Pretraining

You can run the pretraining of model for Imagenet with following code.

torchrun --nproc_per_node=8 main_prosmin.py --arch vit_small --data_path /path/to/imagenet/train --output_dir /path/to/saving_dir

Linear evaluation

You can run the linear-evaluation of the model for Imagenet with the following code.

torchrun --nproc_per_node=8 eval_linear.py --arch vit_small --data_path /path/to/imagenet --pretrained_weights /path/to/saving_dir

Acknowledgement

This repository is built using the DINO repository.

License

This repository is released under the Apache 2.0 license as found in the LICENSE file.