Testing adversarial images against Barrage of Random Transforms Defense method
Original paper:
https://github.com/XttyCTL9/BaRTDefense
This project was built in:
Tensorflow 1.14.0
scipy 1.1.0
scikit-image 0.16.2
Pillow 6.0.0
Checkpoint downloaded from:
https://pan.baidu.com/s/1yAqmqfSODylaZQuF8nZ3vw
There are 10 adversarial images generated under the adversarial_images_resnet50 folder.
These adversarial images are generated using the Trusted-AI adversarial-robustness-toolbox.
The attack method was performed using the HopSkipJump attack on resnet_v2_50 using the untargeted attack mode.