Introduction

Code for ICCV 2023 oral paper "Advancing Example Exploitation Can Alleviate Critical Challenges in Adversarial Training".

Requirements

1. Pytorch
2. Numpy
3. tqdm
4. wandb

Train & Test

1. Fill your data path for variable dir_dataset in config.py.
2. You can configure the train/test parameter in train_adv.py through the config variable, including:
   • Choose the AT method.
   • Set the hyper-parameters.
3. Directly running train_adv.py will perform the training and evaluating process and save the results.