Pytorch | 利用IE-FGSM针对CIFAR10上的ResNet分类器进行对抗攻击

CIFAR数据集
IE-FGSM介绍
- 算法流程
IE-FGSM代码实现
- IE-FGSM算法实现
- 攻击效果
代码汇总

本篇文章我们使用Pytorch实现IE-FGSM对CIFAR10上的ResNet分类器进行攻击.

CIFAR数据集

CIFAR-10数据集是由加拿大高级研究所（CIFAR）收集整理的用于图像识别研究的常用数据集，基本信息如下：

数据规模：该数据集包含60,000张彩色图像，分为10个不同的类别，每个类别有6,000张图像。通常将其中50,000张作为训练集，用于模型的训练；10,000张作为测试集，用于评估模型的性能。
图像尺寸：所有图像的尺寸均为32×32像素，这相对较小的尺寸使得模型在处理该数据集时能够相对快速地进行训练和推理，但也增加了图像分类的难度。
类别内容：涵盖了飞机（plane）、汽车（car）、鸟（bird）、猫（cat）、鹿（deer）、狗（dog）、青蛙（frog）、马（horse）、船（ship）、卡车（truck）这10个不同的类别，这些类别都是现实世界中常见的物体，具有一定的代表性。

下面是一些示例样本：

IE-FGSM介绍

IE-FGSM（Enhanced Euler's Method - Fast Gradient Sign Method）算法是一种基于增强欧拉方法的对抗样本生成算法，旨在提高对抗样本的转移性。以下是IE-FGSM算法的详细流程：

算法流程

算法输入与初始化
- 一个带有真实标签 y t r u e y^{true} ytrue 的干净样本 x x x。
- 一个具有损失函数 J J J 的分类器 f f f。
- 最大扰动预算 ε \varepsilon ε，用于限制对抗样本与原始样本之间的差异。
- 迭代次数 T T T，决定算法的迭代步数。
- 令初始对抗样本 x 0 a d v = x x^{adv}_0 = x x0adv=x。
- 计算步长 α = ε T \alpha = \frac{\varepsilon}{T} α=Tε。
- 初始化梯度 g 0 = 0 g_0 = 0 g0=0。
迭代更新过程
- 计算梯度 ：
  - 计算当前对抗样本 x t a d v x^{adv}_t xtadv 关于损失函数 J J J 的梯度 ∇ x J ( x t a d v , y t r u e ) \nabla_x J(x^{adv}_t, y^{true}) ∇xJ(xtadv,ytrue)。
  - 计算当前梯度的归一化值 g p = ∇ x J ( x t a d v , y t r u e ) ∥ ∇ x J ( x t a d v , y t r u e ) ∥ 1 g_p = \frac{\nabla_x J(x^{adv}_t, y^{true})}{\|\nabla_x J(x^{adv}_t, y^{true})\|_1} gp=∥∇xJ(xtadv,ytrue)∥1∇xJ(xtadv,ytrue)。
  - 计算预期对抗样本 x t a d v + g p x^{adv}_t + g_p xtadv+gp 的梯度 ∇ x J ( x t a d v + g p , y t r u e ) \nabla_x J(x^{adv}_t + g_p, y^{true}) ∇xJ(xtadv+gp,ytrue)，并计算其归一化值 g a = ∇ x J ( x t a d v + g p , y t r u e ) ∥ ∇ x J ( x t a d v + g p , y t r u e ) ∥ 1 g_a = \frac{\nabla_x J(x^{adv}_t + g_p, y^{true})}{\|\nabla_x J(x^{adv}_t + g_p, y^{true})\|_1} ga=∥∇xJ(xtadv+gp,ytrue)∥1∇xJ(xtadv+gp,ytrue)。
- 计算搜索方向 ：
  - 对于IE-FGSM，计算搜索方向 ϕ = g p + g a 2 \phi = \frac{g_p + g_a}{2} ϕ=2gp+ga.
- 更新对抗样本 ：
  - 根据搜索方向 ϕ \phi ϕ，计算对抗样本的更新值 x t + 1 a d v = C l i p x ε ( x t a d v + α ⋅ s i g n ( ϕ ) ) x^{adv}_{t+1} = Clip_x^{\varepsilon}(x^{adv}t + \alpha \cdot sign(\phi)) xt+1adv=Clipxε(xtadv+α⋅sign(ϕ))，其中 C l i p x ε ( ⋅ ) Clip_x^{\varepsilon}(\cdot) Clipxε(⋅) 函数用于将更新后的样本限制在原始样本 x x x 的 L ∞ L{\infty} L∞ 范数 ε \varepsilon ε - 球内， s i g n ( ⋅ ) sign(\cdot) sign(⋅) 是符号函数。
算法输出
- 返回最终生成的对抗样本 x a d v = x T a d v x^{adv} = x^{adv}_T xadv=xTadv。

IE-FGSM代码实现

IE-FGSM算法实现

python 复制代码

import torch
import torch.nn as nn

def IE_FGSM(model, criterion, original_images, labels, epsilon, num_iterations=10):
    """
    IE-FGSM (Enhanced Euler's Method - Fast Gradient Sign Method)

    参数:
    - model: 要攻击的模型
    - criterion: 损失函数
    - original_images: 原始图像
    - labels: 原始图像的标签
    - epsilon: 最大扰动幅度
    - num_iterations: 迭代次数
    """
    # 计算步长
    alpha = epsilon / num_iterations
    # 复制原始图像作为初始的对抗样本
    perturbed_images = original_images.clone().detach().requires_grad_(True)

    for _ in range(num_iterations):
        # 前向传播
        outputs = model(perturbed_images)
        loss = criterion(outputs, labels)

        # 反向传播
        model.zero_grad()
        loss.backward()

        # 获取当前对抗样本的梯度
        data_grad = perturbed_images.grad.data
        # 计算当前梯度的归一化值
        gp = data_grad / torch.sum(torch.abs(data_grad), dim=(1, 2, 3), keepdim=True)
        # 计算预期对抗样本的梯度
        perturbed_images_plus_gp = (perturbed_images + gp).detach().requires_grad_(True)
        outputs_plus_gp = model(perturbed_images_plus_gp)
        loss_plus_gp = criterion(outputs_plus_gp, labels)
        model.zero_grad()
        loss_plus_gp.backward()
        ga = perturbed_images_plus_gp.grad.data / torch.sum(torch.abs(perturbed_images_plus_gp.grad.data), dim=(1, 2, 3), keepdim=True)
        # 计算搜索方向
        phi = (gp + ga) / 2
        # 更新对抗样本
        perturbed_images = perturbed_images + alpha * phi.sign()
        # 裁剪对抗样本，使其在原始图像的epsilon范围内
        perturbed_images = torch.clamp(perturbed_images, original_images - epsilon, original_images + epsilon)
        perturbed_images = perturbed_images.detach().requires_grad_(True)

    return perturbed_images

攻击效果

代码汇总

iefgsm.py

python 复制代码

import torch
import torch.nn as nn

def IE_FGSM(model, criterion, original_images, labels, epsilon, num_iterations=10):
    """
    IE-FGSM (Enhanced Euler's Method - Fast Gradient Sign Method)

    参数:
    - model: 要攻击的模型
    - criterion: 损失函数
    - original_images: 原始图像
    - labels: 原始图像的标签
    - epsilon: 最大扰动幅度
    - num_iterations: 迭代次数
    """
    # 计算步长
    alpha = epsilon / num_iterations
    # 复制原始图像作为初始的对抗样本
    perturbed_images = original_images.clone().detach().requires_grad_(True)

    for _ in range(num_iterations):
        # 前向传播
        outputs = model(perturbed_images)
        loss = criterion(outputs, labels)

        # 反向传播
        model.zero_grad()
        loss.backward()

        # 获取当前对抗样本的梯度
        data_grad = perturbed_images.grad.data
        # 计算当前梯度的归一化值
        gp = data_grad / torch.sum(torch.abs(data_grad), dim=(1, 2, 3), keepdim=True)
        # 计算预期对抗样本的梯度
        perturbed_images_plus_gp = (perturbed_images + gp).detach().requires_grad_(True)
        outputs_plus_gp = model(perturbed_images_plus_gp)
        loss_plus_gp = criterion(outputs_plus_gp, labels)
        model.zero_grad()
        loss_plus_gp.backward()
        ga = perturbed_images_plus_gp.grad.data / torch.sum(torch.abs(perturbed_images_plus_gp.grad.data), dim=(1, 2, 3), keepdim=True)
        # 计算搜索方向
        phi = (gp + ga) / 2
        # 更新对抗样本
        perturbed_images = perturbed_images + alpha * phi.sign()
        # 裁剪对抗样本，使其在原始图像的epsilon范围内
        perturbed_images = torch.clamp(perturbed_images, original_images - epsilon, original_images + epsilon)
        perturbed_images = perturbed_images.detach().requires_grad_(True)

    return perturbed_images

train.py

python 复制代码

import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
from models import ResNet18


# 数据预处理
transform_train = transforms.Compose([
    transforms.RandomCrop(32, padding=4),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])

transform_test = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])

# 加载Cifar10训练集和测试集
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=False, transform=transform_train)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=128, shuffle=True, num_workers=2)

testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=False, transform=transform_test)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)

# 定义设备（GPU或CPU）
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# 初始化模型
model = ResNet18(num_classes=10)
model.to(device)

# 定义损失函数和优化器
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

if __name__ == "__main__":
    # 训练模型
    for epoch in range(10):  # 可以根据实际情况调整训练轮数
        running_loss = 0.0
        for i, data in enumerate(trainloader, 0):
            inputs, labels = data[0].to(device), data[1].to(device)

            optimizer.zero_grad()

            outputs = model(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()

            running_loss += loss.item()
            if i % 100 == 99:
                print(f'Epoch {epoch + 1}, Batch {i + 1}: Loss = {running_loss / 100}')
                running_loss = 0.0

    torch.save(model.state_dict(), f'weights/epoch_{epoch + 1}.pth')
    print('Finished Training')

advtest.py

python 复制代码

import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
from models import *
from attacks import *
import ssl
import os
from PIL import Image
import matplotlib.pyplot as plt

ssl._create_default_https_context = ssl._create_unverified_context

# 定义数据预处理操作
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.491, 0.482, 0.446), (0.247, 0.243, 0.261))])

# 加载CIFAR10测试集
testset = torchvision.datasets.CIFAR10(root='./data', train=False,
                                       download=False, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=128,
                                         shuffle=False, num_workers=2)

# 定义设备（GPU优先，若可用）
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = ResNet18(num_classes=10).to(device)

criterion = nn.CrossEntropyLoss()

# 加载模型权重
weights_path = "weights/epoch_10.pth"
model.load_state_dict(torch.load(weights_path, map_location=device))


if __name__ == "__main__":
    # 在测试集上进行FGSM攻击并评估准确率
    model.eval()  # 设置为评估模式
    correct = 0
    total = 0
    epsilon = 16 / 255  # 可以调整扰动强度
    for data in testloader:
        original_images, labels = data[0].to(device), data[1].to(device)
        original_images.requires_grad = True
        
        attack_name = 'IE-FGSM'
        if attack_name == 'FGSM':
            perturbed_images = FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'BIM':
            perturbed_images = BIM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'MI-FGSM':
            perturbed_images = MI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'NI-FGSM':
            perturbed_images = NI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'PI-FGSM':
            perturbed_images = PI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'VMI-FGSM':
            perturbed_images = VMI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'VNI-FGSM':
            perturbed_images = VNI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'EMI-FGSM':
            perturbed_images = EMI_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'AI-FGTM':
            perturbed_images = AI_FGTM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'I-FGSSM':
            perturbed_images = I_FGSSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'SMI-FGRM':
            perturbed_images = SMI_FGRM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'VA-I-FGSM':
            perturbed_images = VA_I_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'PC-I-FGSM':
            perturbed_images = PC_I_FGSM(model, criterion, original_images, labels, epsilon)
        elif attack_name == 'IE-FGSM':
            perturbed_images = IE_FGSM(model, criterion, original_images, labels, epsilon)
        
        perturbed_outputs = model(perturbed_images)
        _, predicted = torch.max(perturbed_outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

    accuracy = 100 * correct / total
    # Attack Success Rate
    ASR = 100 - accuracy
    print(f'Load ResNet Model Weight from {weights_path}')
    print(f'epsilon: {epsilon:.4f}')
    print(f'ASR of {attack_name} : {ASR :.2f}%')