Yinpeng Dong
Yinpeng Dong
Hi, I used the code at [https://github.com/tensorflow/cleverhans/blob/master/examples/nips17_adversarial_competition/dev_toolkit/run_attacks_and_defenses.py](https://github.com/tensorflow/cleverhans/blob/master/examples/nips17_adversarial_competition/dev_toolkit/run_attacks_and_defenses.py) to generate the `target_class.csv`. Or you could generate this file yourself.
Sorry, where does the error occur?
Hi, Sorry for the late reply. What is the threat model used in your experiments? If you choose the L-infty norm threat model, after you apply the sign gradient, the...
1. Yes, we used the mirror augmentation. But we did not use crop since the original data is 32 * 32. 2. We save the 32bit model.
I think 'current_params' should be saved, not the 'best_p'. Because 'best_p' is only about error.
Sorry I forgot the parameter for ``reg-weight``. After a quick glance of the code, I think it should be 300 instead of 30 since there are 100 classes on CIFAR-100....