debangli / one-pixel-attack-pytorch Goto Github PK

李老师您好：
单像素攻击的论文中分别进行了目标攻击he非目标攻击。我想请教一下您，目标攻击的实现原理。只修改一个像素是如何使模型错误地预测对抗性图像为特定的标签呢？
希望您能百忙之中抽出时间回答我的问题，谢谢！

请问您的代码可以实现在迭代过程中像素点的颜色不变吗

Hello, I got the following error while running the attack. How can I solve it, please

File "attack.py", line 181, in
main()
File "attack.py", line 176, in main
results = attack_all(net, testloader, pixels=args.pixels, targeted=args.targeted, maxiter=args.maxiter, popsize=args.popsize, verbose=args.verbose)
File "attack.py", line 140, in attack_all
flag, x = attack(input, target[0], net, target_calss, pixels=pixels, maxiter=maxiter, popsize=popsize, verbose=verbose)
File "attack.py", line 89, in attack
inits = np.zeros([popmul*len(bounds), len(bounds)])
TypeError: 'float' object cannot be interpreted as an integer

I trained resnet101 and vgg16 without changing the codes but did not get the reported accuracies. Mine is just slightly above 80.