Skip to content

ckomaki/kaggle-nips-2017

Repository files navigation

Submissions for Kaggle nips 2017 adversarial image competition.

How to run Attack and Targeted Attack

  1. Run cleverhans's weight downloader.
  2. Copy corresponding weights to attack/fgsm_tweak_dual and targeted_attack/iter_target_class_tweak_dual.

How to run Defense

  1. Download keras and h5py .whl files into defense/resnet_xception_vgg19_dual
  2. Download corresponding keras pre-trained weights and put them into defense/resnet_xception_vgg19_dual/common.
  3. Somehow fine-tune the above pre-trained weights with adversarial images, and put them under defense/resnet_xception_vgg19_dual.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published