pytorch1.7教程实验——对抗示例生成FGSM

教程原网址：https://pytorch.apachecn.org/#/docs/1.7/21
我们将通过图像分类器上的示例来探讨该主题。具体而言，我们将使用最流行的一种攻击方法，即快速梯度符号攻击（FGSM）来欺骗 MNIST 分类器。

跑的代码需要用到的数据集和预训练模型自行下载，最好去官网下载。

跑通的代码如下：

from __future__ import print_function
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
import numpy as np
import matplotlib.pyplot as pltepsilons = [0, .05, .1, .15, .2, .25, .3]
pretrained_model = "../data/lenet_mnist_model.pth"
use_cuda=True# LeNet Model definition
class Net(nn.Module):def __init__(self):super(Net, self).__init__()self.conv1 = nn.Conv2d(1, 10, kernel_size=5)self.conv2 = nn.Conv2d(10, 20, kernel_size=5)self.conv2_drop = nn.Dropout2d()self.fc1 = nn.Linear(320, 50)self.fc2 = nn.Linear(50, 10)def forward(self, x):x = F.relu(F.max_pool2d(self.conv1(x), 2))x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))x = x.view(-1, 320)x = F.relu(self.fc1(x))x = F.dropout(x, training=self.training)x = self.fc2(x)return F.log_softmax(x, dim=1)# MNIST Test dataset and dataloader declaration
test_loader = torch.utils.data.DataLoader(datasets.MNIST('../data', train=False, download=True, transform=transforms.Compose([transforms.ToTensor(),])),batch_size=1, shuffle=True)# Define what device we are using
print("CUDA Available: ",torch.cuda.is_available())
device = torch.device("cuda" if (use_cuda and torch.cuda.is_available()) else "cpu")# Initialize the network
model = Net().to(device)# Load the pretrained model
model.load_state_dict(torch.load(pretrained_model, map_location='cpu'))# Set the model in evaluation mode. In this case this is for the Dropout layers
model.eval()# FGSM attack code
def fgsm_attack(image, epsilon, data_grad):# Collect the element-wise sign of the data gradientsign_data_grad = data_grad.sign()# Create the perturbed image by adjusting each pixel of the input imageperturbed_image = image + epsilon*sign_data_grad# Adding clipping to maintain [0,1] rangeperturbed_image = torch.clamp(perturbed_image, 0, 1)# Return the perturbed imagereturn perturbed_imagedef test( model, device, test_loader, epsilon ):# Accuracy countercorrect = 0adv_examples = []# Loop over all examples in test setfor data, target in test_loader:# Send the data and label to the devicedata, target = data.to(device), target.to(device)# Set requires_grad attribute of tensor. Important for Attackdata.requires_grad = True# Forward pass the data through the modeloutput = model(data)init_pred = output.max(1, keepdim=True)[1] # get the index of the max log-probability# If the initial prediction is wrong, dont bother attacking, just move onif init_pred.item() != target.item():continue# Calculate the lossloss = F.nll_loss(output, target)# Zero all existing gradientsmodel.zero_grad()# Calculate gradients of model in backward passloss.backward()# Collect datagraddata_grad = data.grad.data# Call FGSM Attackperturbed_data = fgsm_attack(data, epsilon, data_grad)# Re-classify the perturbed imageoutput = model(perturbed_data)# Check for successfinal_pred = output.max(1, keepdim=True)[1] # get the index of the max log-probabilityif final_pred.item() == target.item():correct += 1# Special case for saving 0 epsilon examplesif (epsilon == 0) and (len(adv_examples) < 5):adv_ex = perturbed_data.squeeze().detach().cpu().numpy()adv_examples.append( (init_pred.item(), final_pred.item(), adv_ex) )else:# Save some adv examples for visualization laterif len(adv_examples) < 5:adv_ex = perturbed_data.squeeze().detach().cpu().numpy()adv_examples.append( (init_pred.item(), final_pred.item(), adv_ex) )# Calculate final accuracy for this epsilonfinal_acc = correct/float(len(test_loader))print("Epsilon: {}\tTest Accuracy = {} / {} = {}".format(epsilon, correct, len(test_loader), final_acc))# Return the accuracy and an adversarial examplereturn final_acc, adv_examplesaccuracies = []
examples = []# Run test for each epsilon
for eps in epsilons:acc, ex = test(model, device, test_loader, eps)accuracies.append(acc)examples.append(ex)plt.figure(figsize=(5,5))
plt.plot(epsilons, accuracies, "*-")
plt.yticks(np.arange(0, 1.1, step=0.1))
plt.xticks(np.arange(0, .35, step=0.05))
plt.title("Accuracy vs Epsilon")
plt.xlabel("Epsilon")
plt.ylabel("Accuracy")
plt.show()# Plot several examples of adversarial samples at each epsilon
cnt = 0
plt.figure(figsize=(8,10))
for i in range(len(epsilons)):for j in range(len(examples[i])):cnt += 1plt.subplot(len(epsilons),len(examples[0]),cnt)plt.xticks([], [])plt.yticks([], [])if j == 0:plt.ylabel("Eps: {}".format(epsilons[i]), fontsize=14)orig,adv,ex = examples[i][j]plt.title("{} -> {}".format(orig, adv))plt.imshow(ex, cmap="gray")
plt.tight_layout()
plt.show()

运行结果：

D:\code\pytorch\fsgm-gan
(pytorch) λ python fsgm-gan.py
CUDA Available:  True
Epsilon: 0      Test Accuracy = 9810 / 10000 = 0.981
Epsilon: 0.05   Test Accuracy = 9426 / 10000 = 0.9426
Epsilon: 0.1    Test Accuracy = 8510 / 10000 = 0.851
Epsilon: 0.15   Test Accuracy = 6826 / 10000 = 0.6826
Epsilon: 0.2    Test Accuracy = 4301 / 10000 = 0.4301
Epsilon: 0.25   Test Accuracy = 2082 / 10000 = 0.2082
Epsilon: 0.3    Test Accuracy = 869 / 10000 = 0.0869

pytorch1.7教程实验——对抗示例生成FGSM相关推荐

pytorch1.7教程实验——DCGAN生成对抗网络
教程原网址:https://pytorch.apachecn.org/#/docs/1.7/22 DCGAN 是上述 GAN 的直接扩展,不同之处在于,DCGAN 分别在判别器和生成器中分别使用卷积和 ...
PyTorch 1.0 中文官方教程：对抗性示例生成
译者:cangyunye 作者: Nathan Inkawhich 如果你正在阅读这篇文章,希望你能理解一些机器学习模型是多么有效.现在的研究正在不断推动ML模型变得更快.更准确和更高效.然而,在设计 ...
pytorch1.7教程实验——分类器训练
近来想大致总结一下自己知识学习的脉络,发现自己除了大量的工程经验外,对模型算法的研究还是不够深入,而且大多都是关于目标检测方向,锚框或非锚框以及transformer,其他的涉猎不足,认识不够清晰.而 ...
pytorch1.7教程实验——使用 TensorBoard 可视化模型，数据和训练
可视化还是很重要的,作为官方教程的开头部分,还是有必要好好看看,毕竟使用服务器没有桌面,不用直接使用画图函数看图像,使用tensorboard可以作为一种可视化方法,而且训练过程的损失曲线等也可以通过 ...
pytorch1.7教程实验——迁移学习训练卷积神经网络进行图像分类
只是贴上跑通的代码以供参考学习参考网址:迁移学习训练卷积神经网络进行图像分类需要用到的数据集下载网址: https://download.pytorch.org/tutorial/hymenopt ...
pytorch生成对抗示例
pytorch生成对抗示例本文对ML(机器学习)模型的安全漏洞的认识,并将深入了解对抗性机器学习的热门话题.图像添加难以察觉的扰动会导致模型性能大不相同.通过图像分类器上的示例探讨该主题.使用第一种 ...
动量迭代式对抗噪声生成方法 | VALSE2018年度杰出学生论文奖
编者按:深度模型的精度和速度长期以来成为了评价模型性能的核心标准,但即使性能优越的深度神经网络也很容易被对抗样本攻击.因此,寻找到合适的对抗攻击策略可有效提升模型本身的鲁棒性.本文作者提出了基于动量的 ...
GAN 系列的探索与pytorch实现 (数字对抗样本生成)
GAN 系列的探索与pytorch实现 (数字对抗样本生成) 文章目录 GAN 系列的探索与pytorch实现 (数字对抗样本生成) GAN的简单介绍生成对抗网络 GAN 的基本原理大白话版本非 ...

pytorch1.7教程实验——对抗示例生成FGSM

pytorch1.7教程实验——对抗示例生成FGSM相关推荐

最新文章

热门文章