This repository contains the demo code of the method called REGroup proposed in the paper: REGroup: Rank-aggregating Ensemble of Generative Classifiers for Robust Predictions, IEEE/CVF WACV, 2022.

Requirements

• Pytorch
• numpy, scipy
• matplotlib
• Jupyter notebook
• foolbox (version 2.3.0)

Steps to run the demo

• Clone the repository.
• Download CIFAR10 PGD L-infinity adversarial examples
• Open jupyter notebook REGroup_demo_cifar10_vgg19.ipynb

$ git clone https://github.com/lokender/REGroup.git
$ cd REGroup
$ wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1ylJctBJzh4ih-0zzD4ZLO2umh--QpX7u' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1ylJctBJzh4ih-0zzD4ZLO2umh--QpX7u" -O cifar10_vgg19_pgd_examples.mat && rm -rf /tmp/cookies.txt

To-dos?

• [Done] Classifier: VGG19, Dataset: CIFAR10 ( Released )
• [To-do] Classifier: VGG19, Dataset: ImageNet ( Will be released soon )
• [To-do] Classifier: ResNet, Dataset: CIFAR10 ( Will be released soon )
• [To-do] Classifier: ResNet, Dataset: ImageNet ( Will be released soon )
• [To-do] Classifier: Inception, Dataset: ImageNet ( Will be released soon )
• [To-do] Code for building generative classifiers. ( Will be released soon )

Report any bug or suggestion to [email protected].

Press STAR on the top right of this page for continuous updates.