FaceGSM: Targeted Adversarial Attack using FGSM Method
FaceGSM
FaceGSM designed for performing targeted adversarial attacks using the FGSM (Fast Gradient Sign Method) in Facial Recognition Embedding Model. FaceGSM revolutionizes security testing with a suite of innovative features, including:
- Static – Takes static images as input for FaceGSM.
- Capture – Takes image captured by camera as input for FaceGSM
- Live – Takes real-time live video feed frames as input for FaceGSM.
| Attacker’s Face (Clario) |
Target’s Face (Clints) |
Output : Generated Adversarial Image |
Attack Result : Attacker’s Face Predicted as Victim |
 |
 |
 |
 |
FaceGSM utilizes the FGSM approach to create a subtle layer of semi-invincible pixels. When applied to an image of a person’s face, this layer will make the model to misidentify the face as someone else.
What the name does not imply, however, is that you don’t even need to know what an FGSM is to exploit a facial recognition model using FaceGSM framework. With just access to a facial recognition model and the target’s face of your choice, FaceGSM will attempt to understand the construction of the model, apply image pre-processing accordingly, and then generate layers of perturbation pixels that could make a facial recognition model to misclassify your face into your target’s face.
Key Features
✅ Fully compatible with multiple facial recognition embedding model including FaceNet, ArcFace, GhostFaceNet, DeepID, and VGGFace2
✅ Supports multiple input media, including static image, captured image and live video feed
✅ Saved generated adversarial image as checkpoints to increase efficiency for future attacks
✅ Works with your own Custom Face Datasets
✅ Provide easy installation and intuitive UI/UX
