This repository contains the source code of "Minimizing Adversarial Training Samples for Robust Image Classifiers: Analysis and adversarial example Generator Design".
The repository consists of the following:
adversarial_train.py
: Code for adversarial training in the method.test_acc.py
: Code for performing predictions using the ARF method.generate_and_test.py
: Code for adversarial attack.- 'new_gan/model/wgan_C_gan.py': Code for new GAN network.
- 'new_gan/train_gan.py': Code for training new GAN.
To use the ARF method and comparison methods, follow the instructions below:
- Navigate to the
Minimizing-AT-samples-for-robustness/
directory. - Run
adversarial_train.py
to implement adversarial training. - Use
test_acc.py
to perform predictions using the ARF method. - Use 'new_gan/train_gan.py' to implement GAN training.