/Targeted-Adversarial-Example-Generator

Adversarial example targeted misclassification (iterative least likely class method)

Primary LanguagePythonMIT LicenseMIT

Targeted Adversarial Example Generator

1

Create an adversarial example for targeted misclassification using iterative least likely class method.
The target neural network is Keras MobilenetV2 ImageNet.

Instructions

Requirements

  • Python 3+
  • tensorflow
  • numpy
  • matplotlib

Usage

Input image name : input.jpg
Output image name : output.jpg
main.py and input.jpg must be in the same path.

Run main.py
While main is running, original image, noise, adversarial example are displayed in screen.

You can change a target of misclassification by changing target_index value