AdvMask-A-Sparse-Adversarial-Attack-Based-Data-Augmentation-Method-for-Image-Classification

This is the official implementation of AdvMask, as was used for the paper.

You can directly start off using our implementations on CIFAR-10 and CIFAR-100.

Use AdvMask for data augmentation

• Clone this directory and cd into it.

git clone https://github.com/Jackbrocp/AdvMask

cd AdvMask

Updates

• 2023/7/10: Initial release

Getting Started

Requirements

• Python 3
• PyTorch 1.6.0
• Torchvision 0.7.0
• Numpy

Train Examples

Download the Attack Masks

CIFAR-10

CIFAR-100

Download the attack mask and put them into ./Attack_Mask/.

Parameters

--conf，path to the config file, e.g., confs/resnet18.yaml

Examples

Apply AdvMask as a data augmentation method training ResNet18 model on CIFAR10/100 dataset.

python train.py --conf confs/resnet18.yaml

More Examples

Run additional comparisons on AdvMask combined with other data augmentation methods. (e.g., "AdvMask+AutoAugment") First change mask parameter in the config file, e.g. "AutoAugment", "Fast-AutoAugment"

python additional_comparison.py --conf confs/resnet18.yaml

Citation

If you find this repository useful in your research, please cite our paper:

@article{YANG2023109847, title = {AdvMask: A sparse adversarial attack-based data augmentation method for image classification}, journal = {Pattern Recognition}, volume = {144}, pages = {109847}, year = {2023}, issn = {0031-3203}, doi = {https://doi.org/10.1016/j.patcog.2023.109847}, url = {https://www.sciencedirect.com/science/article/pii/S0031320323005459}, author = {Suorong Yang and Jinqiao Li and Tianyue Zhang and Jian Zhao and Furao Shen} }