erhul issues

Results 3 issues of


                                            erhul

Confusion about L2-norm normalization in PGD attack

``` elif norm == "l_2": g_norm = torch.norm(g.view(g.shape[0],-1),dim=1).view(-1,1,1,1) scaled_g = g/(g_norm + 1e-10) d = (d + scaled_g*alpha).view(d.size(0),-1).renorm(p=2,dim=0,maxnorm=epsilon).view_as(d) ``` Why is the `renorm` operation in the L2-norm PGD attack, i.e.,...

The problem of gradient overflow during ImageNet training

![image](https://github.com/user-attachments/assets/2ec082b4-f6d4-4bae-a40c-8e465bf54d56) Why does gradient overflow occur? It seems that it doesn't affect the experimental results. Have you encountered this situation during training? How did you solve it?

Question about the differentiability of self-supervised augmentations in the paper "Adversarial Attacks are Reversible with Natural Supervision"

from torchvision.transforms import transforms # color_jitter = transforms.ColorJitter(0.8 * s, 0.8 * s, 0.8 * s, 0.2 * s) transforms = torch.nn.Sequential( transforms.RandomResizedCrop(size=size), transforms.RandomHorizontalFlip(), # transforms.RandomApply([color_jitter], p=0.8), transforms.ColorJitter(0.8 * s,...