The code for CVPR2019 (ComDefend: An Efficient Image Compression Model to Defend Adversarial Examples) paper The pure tensorflow of code is released in the link

tensorflow>=1.1
python3
canton(pip install canton)

The keras and pytorch of the code will be released soon.

clean_image: we select 7 clean images which are classified correctly by the classifier
attack_image: we select 7 adversarial images which are attacked by the FGSM method
defend_image: we use the Comdefend to deal with 7 adversarial images
chackpoints: the model parameters
com_imagenet_temp, temp_imagenet: the temporary files of the Comdefend
dev.csv: correspondence between images and labels
Resnet_imagenet.py: the classifier
compression_imagenet.py: the Comdefend for Imagenet
compression_mnist.py: the Comdefend for fashion_mnist

E-mail: [email protected] or [email protected]