AutoView

Learning Self-Regularized Adversarial Views for Self-Supervised Vision Transformers
Tao Tang∗, Changlin Li∗, Guangrun Wang, Kaicheng Yu, Xiaojun Chang, Xiaodan Liang^†

(*: equal contribution, †: corresponding author)

[arXiv Preprint]

Introduction

We propose AutoView, a self-regularized adversarial AutoAugment method, to learn views for self-supervised vision transformers.

First, we reduce the search cost of AutoView to nearly zero by learning views and network parameters simultaneously in a single forward-backward step, minimizing and maximizing the mutual information among different augmented views, respectively.
Then, to avoid information collapse caused by the lack of label supervision, we propose a self-regularized loss term to guarantee the information propagation.
Additionally, we present a curated augmentation policy search space for self-supervised learning, by modifying the generally used search space designed for supervised learning.
On ImageNet, our AutoView achieves remarkable improvement over RandAug baseline (+10.2% k-NN accuracy), and consistently outperforms sota manually tuned view policy by a clear margin. Extensive experiments show that AutoView pretraining also benefits downstream tasks and improves model robustness.

Visualization

Getting Started

git clone https://github.com/Trent-tangtao/AutoView.git

This is a preliminary release. We have not carefully organized everything now.

Citation

If you find AutoView is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@article{tang2022learning,
  title={Learning Self-Regularized Adversarial Views for Self-Supervised Vision Transformers}, 
  author={Tao Tang and Changlin Li and Guangrun Wang and Kaicheng Yu and Xiaojun Chang and Xiaodan Liang},
  journal={arXiv preprint arXiv:2210.08458},
  year={2022}
}