Questions about indicator calculation and adversarial training

[wyattxuanyang]

Thank you for releasing the code. I've got two questions：

1. I wonder whether there's code for the calculation of Robust Accuracy in standard training, since it was demonstrated in the paper but I didn't find it in train_standard,py (Standard Accuracy only) Or rather, how to calculate Robust Accuracy for Resnets？
2. How can I conduct adversarial training? Are there codes or i just need to add Gaussian noise on the input data？