search

csdongxian / skip-connections-matter

Codes for ICLR 2020 paper "Skip Connections Matter: On the Transferability of Adversarial Examples Generated with ResNets"
MIT License
70 stars 9 forks source link

Code for tensorflow version？ #7

Closed mesunhlf closed 2 years ago

mesunhlf commented 2 years ago

Hi, this is a insteresting work! I find that the repo is torch version. Can you offer me the Tensorflow code of SGM, thanks!

mesunhlf commented 2 years ago

I use torch code to generate adversarial examples on Resnet50 for fooling adv trained I3ens3 or IR2ens (tf. version). However, I find the performance is lower than the results in the paper. I am wondering whether I need to run TensorFlow version of SGM?

csdongxian commented 2 years ago

Hi, thanks for your interest in our paper.

I find the performance is lower than the results in the paper

  1. Do you select the appropriate hyper-parameter \gamma? The optimal hyper-parameter \gamma for SGM on ResNet should be gamma=0.2 (a quite low value).
  2. As an additional question, can the attack success rate of SGM be higher than that of the PGD method?

Can you offer me the Tensorflow code of SGM

I don't have the code for the Tensorflow version. Here I can provide my own generated adversarial samples. You can download them from Google drive or Baidu drive (pwd: je7x), and evaluate SGM performance based on them.

Hope this can be helpful :)

mesunhlf commented 2 years ago

您的邮件我已收到，O(∩_∩)O谢谢

csdongxian commented 2 years ago

If you have more questions, be free to send an email or reopen this issue.