The rapid development of Deep Learning (DL) and, more specifically, Convolutional Neural Networks (CNNs) has achieved high accuracy over the past decade, becoming the standard approach in computer vision in a short time. However, recent studies have discovered that CNNs are vulnerable to adversarial attacks in image classification tasks. While most studies have focused on DL models for image classification, only a few works have addressed this issue in the context of Image Quality Assessment (IQA). This paper investigates the robustness of different CNN models against adversarial attacks when used for an IQA task. We propose an adaptation of state-of-the- art image classification attacks in both targeted and untargeted modes for an IQA regression task. We also analyze the correlation between the perturbation’s visibility and the attack’s success. Our experimental results show that DL-based IQA methods are vulnerable to such attacks, with a significant decrease in correlation scores when subjected to adversarial perturbations. Consequently, the development of countermeasures against such attacks is essential for improving the reliability and accuracy of DL-based IQA models.
AttacksGenerationOnDataset.ipynb provides a guideline on how to launch the attack on a dataset.
- Set the path to the target dataset's repository
db = "../Databases/tid2013/" - Define the maximum scale of the ground truth quality score:
scale = int(input()) - Define the victim model and load its weights
model = initialize_model('inception',False,True)weights_path = '../pretrained/iqaModel_tid_inception.pth' - Define the parametters of the attack
iterations = [10]epsilons = [0.001,0.01,0.1]attacks = ["bim","pgd","fgm"]losses = ['mse(y_tielda,y)']
We kindly ask you to cite our paper if you find the repository useful to your work:
@inproceedings{meftah2023evaluating,
title={Evaluating the Vulnerability of Deep Learning-based Image Quality Assessment Methods to Adversarial Attacks},
author={Meftah, Hanene FZ Brachemi and Fezza, Sid Ahmed and Hamidouche, Wassim and D{\'e}forges, Olivier},
booktitle={2023 11th European Workshop on Visual Information Processing (EUVIP)},
pages={1--6},
year={2023},
organization={IEEE}
}
This project is funded by both Région Bretagne (Brittany region), France, CREACH Labs and Direction Générale de l’Armement (DGA). We also used the code provided by the CleverHans software library to which we added further modifications in order to adapt it to the context of our study.
Hanene F.Z Brachemi Meftah , [email protected]
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent