columbia / pixeldp Goto Github PK

in carlini.py line 366 I guess it should suppose to be :
softmax_predictions = self.sess.run(self.predictions, args)
argmax_predictions = np.argmax(softmax_predictions, axis=1)

I wonder what the difference between carlini.py and carlini_robust_precision.py in the code, they look similar.

Hi, the file models/utils/utils.py seems to be missing

When I ran code python3 main.py --num_gpus 1 --dataset mnist --mode plot , I got the warning and nothing image was shown on the screen.
UserWarning: This figure includes Axes that are not compatible with tight_layout, so results might be incorrect. warnings.warn("This figure includes Axes that are not compatible "
Please help.

The while loop in line 352 of carlini.py never exits. :(

Some code bugs:

1. Can the variable "d" be removed safely in line 89 of experiments/cifar10_robust_prec.py?
2. 'L_sigmas' is not defined in cifar10_eval.py
3. The variable 'max_attack_size' is not defined in line 190 of experiments/cifar10_robust_prec.py, maybe it should be 'attack_size'?

Addition:
Could you give more detailed annotations about each file in experiments folder? It's not so easy to see which file is used to plot a certain figure in the paper?

Thanks in advance.

There isn't the variable 'max_attack_size' in line 190 of experiments/cifar10_robust_prec.py, maybe it should be 'attack_size'?

Dear author,

Thanks for your brilliant work. After reading your paper, I am interested in the performance of pixeldp_cnn under PGD attack. I notice that you have implemented one in this repo (Just as the paper mentioned in the Appendix). However, It seems that something is missing.

First of all, in the train_attack.py, the initialization of the attack:
attack = attack_class.Attack
is only for the Carlini. However Carlini has different parameters with PGD for initialization.

What's more, in the pgd.py, parameters like
self.min_pix = min_pix
self.max_pix = max_pix
are also missing.