Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
sy_liang 64ef12e303 | 1 year ago | |
---|---|---|
.. | ||
.idea | 1 year ago | |
Asym-AdvCL | 1 year ago | |
Asym-RoCL | 1 year ago | |
apex @ 4ad9b3bd4f | 1 year ago | |
figures | 1 year ago | |
README.md | 1 year ago |
This is a PyTorch implementation of the paper Adversarial Contrastive Learning via Asymmetric InfoNCE (ECCV 2022).
In each sub-directory (./Asym-AdvCL
, ./Asym-RoCL
) for each baseline, please first create ./logs
directories, then download CIFAR datasets into ~/data
. All models will be stored in ./checkpoint
, and logs stored in ./logs
.
This repo is a modification on the RoCL and AdvCL repos. Environments can also be installed according to the requirements of RoCL and AdvCL for experiments of each baseline.
We provide shells for reproducing our main results in Table 1. The hyperparameter settings of baselines are the same as their original papers reported. All experiments were conducted on 2 Tesla V100 GPUs.
First, for RoCL:
sh shell/rocl-cifar10.sh rocl-cifar10 ResNet18 cifar-10 | tee logs/rocl-cifar10.log
sh shell/rocl-IP-cifar10.sh rocl-IP-cifar10 ResNet18 cifar-10 | tee logs/rocl-IP-cifar10.log
sh shell/rocl-HN-cifar10.sh rocl-HN-cifar10 ResNet18 cifar-10 | tee logs/rocl-HN-cifar10.log
sh shell/rocl-IPHN-cifar10.sh rocl-IPHN-cifar10 ResNet18 cifar-10 | tee logs/rocl-IPHN-cifar10.log
sh shell/rocl-cifar100.sh rocl-cifar100 ResNet18 cifar-100 | tee logs/rocl-cifar100.log
sh shell/rocl-IP-cifar100.sh rocl-IP-cifar100 ResNet18 cifar-100 | tee logs/rocl-IP-cifar100.log
sh shell/rocl-HN-cifar100.sh rocl-HN-cifar100 ResNet18 cifar-100 | tee logs/rocl-HN-cifar100.log
sh shell/rocl-IPHN-cifar100.sh rocl-IPHN-cifar100 ResNet18 cifar-100 | tee logs/rocl-IPHN-cifar100.log
For AdvCL:
sh shell/cifar10/advcl-cifar10.sh advcl-cifar10 | tee logs/advcl-cifar10.log
sh shell/cifar10/advcl-IP-cifar10.sh advcl-IP-cifar10 | tee logs/advcl-IP-cifar10.log
sh shell/cifar10/advcl-HN-cifar10.sh advcl-HN-cifar10 | tee logs/advcl-HN-cifar10.log
sh shell/cifar10/advcl-IPHN-cifar10.sh advcl-IPHN-cifar10 | tee logs/advcl-IPHN-cifar10.log
sh shell/cifar100/advcl-cifar100.sh advcl-cifar100 | tee logs/advcl-cifar100.log
sh shell/cifar100/advcl-IP-cifar100.sh advcl-IP-cifar100 | tee logs/advcl-IP-cifar100.log
sh shell/cifar100/advcl-HN-cifar100.sh advcl-HN-cifar100 | tee logs/advcl-HN-cifar100.log
sh shell/cifar100/advcl-IPHN-cifar100.sh advcl-IPHN-cifar100 | tee logs/advcl-IPHN-cifar100.log
If you find our work useful or provides some new insights about adversarial contrastive learning😊, please consider citing:
@inproceedings{yu2022adversarial,
title={Adversarial Contrastive Learning via Asymmetric InfoNCE},
author={Yu, Qiying and Lou, Jieming and Zhan, Xianyuan and Li, Qizhang and Zuo, Wangmeng and Liu, Yang and Liu, Jingjing},
booktitle={European Conference on Computer Vision},
pages={53--69},
year={2022},
organization={Springer}
}
We thank for the code implementation from RoCL, AdvCL, HCL and SupContrast.
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》