sg47 / pqp

Code for "Perceptual Quality-preserving Black-Box Attack against Deep Learning Image Classifiers"

PQP: Perceptual Quality-Preserving black-box attack

Codebase for reproducing the results in the paper "Perceptual Quality-preserving Black-Box Attack against Deep Learning Image Classifiers". The paper can be found on arxiv.

Requirements

• Python3
• TensorFlow 1.14 (may work with previous versions, too).
• To replicate the results of the paper, PyTorch 1.0 is also needed.
• We used the PyTorch models and the weights provided here. Import the python module and download the weights in './CIFAR10_pretrained_models/'.

CIFAR10 test

To run the code with default parameters, type:

python main.py

Modify the main.py file according to your needs. For any problem or comment, do not hesitate to contact me.

Citation

@misc{gragnaniello2019perceptual,
    title={Perceptual Quality-preserving Black-Box Attack against Deep Learning Image Classifiers},
    author={Diego Gragnaniello and Francesco Marra and Giovanni Poggi and Luisa Verdoliva},
    year={2019},
    eprint={1902.07776},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}