Skip to content

MrChenFeng/Adaptive-Soft-Contrastive-Learning_ICPR2022

Repository files navigation

ASCL: Adaptive Soft Contrastive Learning

drawing

This is the official PyTorch implementation of ASCL.

Abstract

Self-supervised learning has recently achieved great success in representation learning without human annotations. The dominant method — that is contrastive learning is generally based on instance discrimination tasks, i.e., individual samples are treated as independent categories. However, presuming all the samples are different contradicts the natural grouping of similar samples in common visual datasets, e.g., multiple views of the same dog. To bridge the gap, this paper proposes an adaptive method that introduces soft inter-sample relations, namely Adaptive Soft Contrastive Learning (ASCL). More specifically, ASCL transforms the original instance discrimination task into a multi-instance soft discrimination task, and adaptively introduces inter-sample relations. As an effective and concise plug-in module for existing self-supervised learning frameworks, ASCL achieves the best performance on several benchmarks in terms of both performance and efficiency.

@INPROCEEDINGS{ASCL,  
author={Feng, Chen and Patras, Ioannis},  
booktitle={2022 26th International Conference on Pattern Recognition (ICPR)},   
title={Adaptive Soft Contrastive Learning},   
year={2022},  
volume={},  
number={},  
pages={2721-2727},  
doi={10.1109/ICPR56361.2022.9956660}
}

Please considering cite our paper and star this repo if you find this repo useful.

Preparation

  • pytorch
  • tqdm
  • wandb

Usage

An example run on CIFAR100 dataset with ASCL model:

python main.py  --epochs 200 --gpuid 0 --dataset cifar100 --data_path data/CIFAR100 --K 5  --type ascl --t1 0.1 --t2 0.05 --aug weak_augment

drawing

ASCL VS MoCo

For users who are not familiar with wandb, please try main_simple.py.

python main_simple.py  --epochs 200 --gpuid 0 --dataset cifar100 --data_path data/CIFAR100 --K 5  --type ascl --t1 0.1 --t2 0.05 --aug weak_augment

For experiments on ImageNet-1K, we simply modified the official MoCo repo. For brevity of this repo, we provide a modified builder.py which can be utilized with the original MoCo repo.

License

This project is licensed under the terms of the MIT license.

Acknowledgement

This work was supported by the EU H2020 AI4Media No. 951911 project.