/EffSSL

Towards Efficient and Effective Self-Supervised Learning of Visual Representations

Primary LanguagePythonMIT LicenseMIT

Efficient and Effective Self-Supervised Learning

This repository contains the implementation of our paper titled "Towards Efficient and Effective Self-Supervised Learning of Visual Representations", accepted at ECCV'22 [Poster][Paper][Video]. A preliminary version of our work was presented at the NeurIPS'21 Workshop, Self-supervised learning, Theory and Practice [Poster][Paper].

plot

Requirements

  • Python 3.8.8
  • PyTorch 1.7.1
  • tqdm

Training and Evaluation

Please check the training scripts provided under the scripts folder to train the base model (SwAV) and ours (SwAV+Rotnet), followed by linear evaluation.

Citation

If you use our code in your research, please cite the following:

@inproceedings{
EffSSL22,
title={Towards Efficient and Effective Self-Supervised Learning of Visual Representations},
author={Sravanti Addepalli and Kaushal Santosh Bhogale and Priyam Dey and Venkatesh Babu Radhakrishnan},
booktitle={European Conference on Computer Vision 2022},
year={2022}
}

Licence

This source code is released under the MIT license, included here.

This project also borrows code from SwAV official repository, also MIT Licensed.