Code and data related WACV 2021 paper: "Defense-friendly Images in Adversarial Attacks: Dataset and Metrics for Perturbation Difficulty" Paper
The dataset contains 3 CSV files with information about the ɛ-robustness levels of the images in the dataset. Easy Images, ɛ-robust and Defense-friendly are terms relative to the attacks and defenses used. However, we provide the ɛ-robustness information of each image in this dataset according to our experiments as a reference.
The dataset is available in the link below:
- Dataset v. 1.0 = Download dataset
If you want to contribute to improve the dataset. Please write an issue and send a Pull Request.
If you use this dataset or any code for your research, please consider citing:
@InProceedings{Pestana_2021_WACV,
author = {Pestana, Camilo and Liu, Wei and Glance, David and Mian, Ajmal},
title = {Defense-Friendly Images in Adversarial Attacks: Dataset and Metrics for Perturbation Difficulty},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
month = {January},
year = {2021},
pages = {556-565}
}
The data and code provided is for research purposes only. No commercial license is provided. For any other questions please contact camilo.pestanacardeno@research.uwa.edu.au