when i run trains_model.py i am getting this error....i am beginner,if any help would be greatfull. about nn_robust_attacks HOT 12 CLOSED

You need to configure keras to use the tensorflow backend. It's currently set to theano.

from nn_robust_attacks.

Where can I get the attack images? Thanks!

from nn_robust_attacks.

They are returned as the output of the attack method.

from nn_robust_attacks.

Thanks! I finished running the files but still don't see any image output. Is there anyway to store the attack images?

from nn_robust_attacks.

You will have to save them with scipy.misc.imsave or other method.

from nn_robust_attacks.

So should I add this command to the show(img) function in the test_attack.py? Thanks

from nn_robust_attacks.

That would work, yeah.

from nn_robust_attacks.

Sorry for bothering again. In line 27 of test_attack.py what is the purose of "if len(img) != 784: return" How should I modify it in order to get adversarial images of Cifar-10?

from nn_robust_attacks.

The show method is useful for MNIST only. You will need to write a new one for CIFAR-10.

from nn_robust_attacks.

Is it necessary to run the train_model.py before running the test_attack.py? Since I am running out of memory

from nn_robust_attacks.

If you don't train the model ahead of time, it doesn't make that much sense to attack it, unless you're trying to investigate some kind of robustness property of the architecture itself using randomly initialized neural networks.

from nn_robust_attacks.

Thanks Carlini. Since my main goal is just to get some cifar10 adversarial image data that can attack some other pre-trained neural network (like Resnet).

from nn_robust_attacks.