深度神经网络已经在图像识别、语音识别等方面取得优秀成果。但是深度神经网络也有它的脆弱性，在面对对抗样本的时候，会出现错误分类，今天将根据FGSM的方法生成相应的对抗样本。

使用的是jupyter notebook。图片自己去下载先，或者替换成自己的图片。。

import torch
import torchvision
from torchvision import datasets,transforms
from torch.autograd import Variable
import torch.utils.data.dataloader as Data
import torch.nn as nn
from torchvision import models
import numpy as np
import cv2
# 导入相应的库之后定义函数

def show_image_diff(original_img,original_label,adversarial_img,adversarial_label):
    # 对比展示原始图片和对抗样本图片
    import matplotlib.pyplot as plt
    plt.figure()
    
    if original_img.any()>1.0:
        original_img=original_img/255.0# 归一化
    if adversarial_img.any()>1.0:
        adversarial_img=adversarial_img/255.0
        
    plt.subplot(131)
    plt.title('Original')
    plt.imshow(original_img)
    plt.axis('off')
    plt.subplot(132)
    plt.title('Adversarial')
    plt.imshow(adversarial_img)
    plt.axis('off')

    plt.subplot(133)
    plt.title('Adversarial-Original')
    difference=adversarial_img-original_img
    # (-1,1)->(0,1)
    difference=difference/abs(difference).max()/2.0+0.5
    plt.axis('off')
    plt.tight_layout()
    plt.imshow()
    

device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')

image_path=r"C:\Users\Lizhu\Downloads\adversarial_examples-master\picture\cow.jpeg"
orig=cv2.imread(image_path)[...,::-1]
orig=cv2.resize(orig,(224,224))
img=orig.copy().astype(np.float32)

mean=[.485,.456, 0.406]
std=[.229,.224,.225]
img/=255.0
img=(img-mean)/std
img=img.transpose(2,0,1)# 将hwc转成chw

img=np.expand_dims(img,axis=0)
img=Variable(torch.from_numpy(img).to(device).float())
print(img.shape)# should be torch.Size([1,3,224,224])
# 使用模型预测
model=models.alexnet(pretrained=True).to(device).eval()
label=np.argmax(model(img).data.cpu().numpy())
print("label{}".format(label))# should be 345

# 设置梯度可获取
img.requires_grad=True

# model的参数不用改变，因此设置梯度为false
for param in model.parameters():
    param.requires_grad = False

optimizer = torch.optim.Adam([img])
loss_func = torch.nn.CrossEntropyLoss()

# 超参数设置
epochs=100
e=0.001
target=266
target=Variable(torch.Tensor([float(target)].to(device).long())# variable定义的变量是可导的

for epoch in range(epochs):
    output=model(img)
    loss = loss_func(output,target)
    label=np.argmax(output.data.cpu().numpy())
    print("epoch={} loss={} label={}".format(epoch,loss,label))
    
    # 如果定向攻击成功
    if label=tatget:
        print("")
        break
    
    # 梯度清零
    optimizer.zero_grad()
    # 反向传递梯度信息
    loss.backward()
    img.data=img.data-e*torch.sign(img.grad.data)# 利用梯度信息更新输入图像x，e对应论文中的namida

adv=img.data.cpu().numpy()[0]
print(adv.shape)# should be (3,224,224)
adv =adv.transpose(1,2,0)# chw->hwc
adv=(adv*std)+mean
adv=adv*255.0
adv=np.clip(adv,0,255).astype(np.uint8)# np.clip()函数，将数值限定在一个范围内，此处，限制为0~255，同torch.clamp()
# 展示图像和不同
show_img_diff(orig,388,adv,target.data.cpu().numpy()[0])