Adversarial Transformation Network(ATN)

Introduction

A simple implement of an Adversarial Autoencoding ATN(AAE ATN) proposed in Adversarial Transformation Networks: Learning to Generate Adversarial Examples using tensorflow.

python 3.5

tensorflow 1.1.0

matplotlib (for result visualizing)

You can test with my trained model:

python atn.py

If you want to train by yourself:

python atn.py --train

Here are some visualized samples:

Before attack, the accuracy of the target cnn network is 0.9902, and it becomes 0.2773 after attack.

The result is not good enough, so WELCOME CONTRIBUTION !!!