Bridging the Gap from Asymmetry Tricks to Decorrelation Principles in Non-contrastive Self-supervised Learning

Pretrained Model

epochs	batch size	acc1	acc5	download
100	2048	69.0%	88.8%	ResNet-50	full checkpoint	train logs	val logs

Pre-Training

python main.py /path/to/imagenet/

Evaluation: Linear Classification

python evaluate.py /path/to/imagenet/ /path/to/checkpoint/resnet50.pth --lr-classifier 0.3

Citation

@article{liu2022bridging,
  title={Bridging the gap from asymmetry tricks to decorrelation principles in non-contrastive self-supervised learning},
  author={Liu, Kang-Jun and Suganuma, Masanori and Okatani, Takayuki},
  journal={Advances in Neural Information Processing Systems},
  volume={35},
  pages={19824--19835},
  year={2022}
}

Acknowledgement

Our code is inherited from Barlow Twins. We thank the authors of the open source project.