This Git repository contains code implementations related to Adversarial Face Detection and Defense. The repository is organized into three main folders:
The 'face_detection' folder contains the implementation of YOLOv5 for face detection on the Human Face dataset. The folder includes the necessary scripts, configuration files, and dataset loading code to train and evaluate the YOLOv5 model on the Wider Face dataset.
- Detailed usage instructions and setup guide can be found in the 'face_detection' folder.
The 'AdversarialAttack' folder contains code related to adversarial attacks on deep learning models, particularly face detection models. Various attack techniques like FGSM, Adding noise to the face after face detection are implemented to generate adversarial examples for the face detection model.
In the folder AttacksAdversarial, put the image in the folder data/inference_image and run !python3 inference.py
In the folder AttacksAdversarial , run the file fgsm-attack.ipynb
- Detailed usage instructions and setup guide can be found in the 'AdversarialAttack' folder.
The 'AdversarialDefense' folder contains code implementations of adversarial defense techniques for robust face detection.
In the 'AdversarialDefense/AdversarialTraining' subfolder, you'll find the implementation of adversarial training methods to enhance the robustness of the face detection model against adversarial attacks.
The 'AdversarialDefense/Denoising' subfolder contains implementations of denoising techniques to remove noise from adversarial examples before passing them to the face detection model.
In the 'AdversarialDefense/Denoising/DenoisingUsingFilters' subfolder, you'll find implementations of various filtering techniques (e.g., Gaussian, Median) to denoise adversarial examples.
In the 'AdversarialDefense/Denoising/DenoisingUsingSRGAN' subfolder, you'll find the implementation of denoising adversarial examples using SRGAN (Super-Resolution Generative Adversarial Network).
You can run the GUI.ipynb notebook for GUI of Face Defender
You can view our slide and presentation on this link : https://goeco.link/HhJUm
- Detailed usage instructions and setup guides for each defense technique can be found in the respective subfolders.