Official Pytorch implementation for "Transferable Adversarial Attacks on Vision Transformers with Token Gradient Regularization" (CVPR 2023).
25
stars
5
forks
source link
你好,因为代码兼容性问题,我将register_backward_hook修改成了register_full_forward_hook,并对v_tgr的返回进行了一些修改,可是最后执行结果有差异。这是我的v_tgr的返回,以及我的环境和执行结果: #4
Closed
guox18 closed 1 year ago