Tensorflow implementation of Attention-ATN: A Method to Generate Transferable Adversarial Examples
USAGE
Download the weight about defense model,cycle_gan model and based-model used to calculate cam-matrix from Google clouds Download the data sets:AAAC-2019 data sets
First run:
pip install -r requirements.txt
Next, run:
python train.py
After training some steps, the chenkpoints and attack images will be saved to local file holder.
The loss curve of training Attention-ATN:
Compare with raw images and adversarial examples:
 image1 |
 Grad_cam1 |
 adversarial example1 |
 image2 |
 Grad_cam2 |
 adversarial example2 |
 image3 |
 Grad_cam3 |
 adversarial example3 |