GitXplorerGitXplorer
k

atgan

public
2 stars
0 forks
0 issues

Commits

List of commits on branch master.
Unverified
6743353222af8d12b2a3d54b4d473e66e89e9ae2

Merge branch 'master' of https://github.com/kawine/atgan

rrabachi committed 7 years ago
Unverified
c889dbe859e57620ad87a8afe7befb13b0d82b22

more user-friendly test script

rrabachi committed 7 years ago
Verified
6a7c87b31a62e8810e074997d201ba99a4374d5e

Update README.md

rrabachi committed 7 years ago
Verified
fe912b7993d67b6652fc5408b7b6c32a9a3b5751

Update README.md

rrabachi committed 7 years ago
Unverified
e3226592992b982997e26afef801b7ce9300ac59

made test script more user-friendly

rrabachi committed 7 years ago
Verified
16dd78b0cbc0629b28c108fbcd654be64dea3979

Update example.py

rrabachi committed 7 years ago

README

The README file for this repository.

Crafting Adversarial Attacks with Adversarial Transformations

Generative Adversarial Transformation Network (GATN); CSC2541 project. By: Kawin, Joey, Romina, Mohammad

  1. Prerequisites: Pytorch installed from source (https://github.com/pytorch/pytorch#from-source), Python 2.7, Numpy

  2. To train the models, run attacks/example.py. This will save all the models under attacks/saved/ . Pretrained models are available at: https://drive.google.com/open?id=1H_vmsv6H_kQs1f_oi39d58b1UBCcrRmt

  3. To test the accuracies of various trained models with different attacker/discriminator combinations, and compare accuracies to FGSM and Carlini-Wagner attacks, run attacks/test_all.py with the different options (use --help to see the options).

  4. You can read through ape-gan-defense.ipynb to see how we evaluated our results against the defense APE-GAN as implemented by https://github.com/carlini/APE-GAN.