292 lines
12 KiB
Python
292 lines
12 KiB
Python
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
|
|
from scipy import stats
|
|
import matplotlib.pyplot as plt
|
|
import cv2
|
|
from mnist import Net
|
|
from pykuwahara import kuwahara
|
|
|
|
epsilons = np.arange(0.0,0.35,0.05)
|
|
pretrained_model = "mnist_cnn_unfiltered.pt"
|
|
use_cuda=False
|
|
|
|
torch.manual_seed(69)
|
|
|
|
|
|
test_loader = torch.utils.data.DataLoader(
|
|
datasets.MNIST('data/', train=False, download=True, transform=transforms.Compose([
|
|
transforms.ToTensor(),
|
|
transforms.Normalize((0.1307,), (0.3081,)),
|
|
])),
|
|
batch_size=1, shuffle=True)
|
|
|
|
print("CUDA Available: ", torch.cuda.is_available())
|
|
device = torch.device("cuda" if use_cuda and torch.cuda.is_available() else "cpu")
|
|
|
|
model = Net().to(device)
|
|
print(type(model))
|
|
|
|
model.load_state_dict(torch.load(pretrained_model, map_location=device))
|
|
|
|
model.eval()
|
|
|
|
def fgsm_attack(image, epsilon, data_grad):
|
|
# Collect the element-wise sign of the data gradient
|
|
sign_data_grad = data_grad.sign()
|
|
|
|
# Create the perturbed image by adjusting each pixel of the input image
|
|
perturbed_image = image + epsilon*sign_data_grad
|
|
|
|
# Adding clipping to maintain [0, 1] range
|
|
perturbed_image = torch.clamp(perturbed_image, 0, 1)
|
|
|
|
return perturbed_image
|
|
|
|
def denorm(batch, mean=[0.1307], std=[0.3081]):
|
|
"""
|
|
Convert a batch of tensors to their original scale.
|
|
|
|
Args:
|
|
batch (torch.Tensor): Batch of normalized tensors.
|
|
mean (torch.Tensor or list): Man used for normalization.
|
|
std (torch.Tensor or list): Standard deviation used for normalization.
|
|
|
|
Returns:
|
|
torch.Tensor: batch of tensors without normalization applied to them.
|
|
"""
|
|
|
|
if isinstance(mean, list):
|
|
mean = torch.tensor(mean).to(device)
|
|
if isinstance(std, list):
|
|
std = torch.tensor(std).to(device)
|
|
|
|
return batch * std.view(1, -1, 1, 1) + mean.view(1, -1, 1, 1)
|
|
|
|
def test(model, device, test_loader, epsilon):
|
|
# Original dataset correct classifications
|
|
orig_correct = 0
|
|
# Attacked dataset correct classifications
|
|
unfiltered_correct = 0
|
|
kuwahara_correct = 0
|
|
bilateral_correct = 0
|
|
gaussian_blur_correct = 0
|
|
random_noise_correct = 0
|
|
snap_color_correct = 0
|
|
one_bit_correct = 0
|
|
plurality_correct = 0
|
|
|
|
adv_examples = []
|
|
|
|
for data, target in test_loader:
|
|
data, target = data.to(device), target.to(device)
|
|
data.requires_grad = True
|
|
|
|
output_orig = model(data)
|
|
orig_pred = output_orig.max(1, keepdim=True)[1]
|
|
|
|
# Calculate the loss
|
|
loss = F.nll_loss(output_orig, target)
|
|
|
|
# Zero all existing gradients
|
|
model.zero_grad()
|
|
|
|
# Calculate gradients of model in backward pass
|
|
loss.backward()
|
|
|
|
# Collect ''datagrad''
|
|
data_grad = data.grad.data
|
|
|
|
# Restore the data to its original scale
|
|
data_denorm = denorm(data)
|
|
|
|
# Apply the FGSM attack
|
|
perturbed_data = fgsm_attack(data_denorm, epsilon, data_grad)
|
|
|
|
# Reapply normalization
|
|
perturbed_data_normalized = transforms.Normalize((0.1307,), (0.3081,))(perturbed_data)
|
|
|
|
# Filter the attacked image
|
|
kuwahara_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="kuwahara")
|
|
bilateral_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="bilateral")
|
|
gaussian_blur_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="gaussian_blur")
|
|
random_noise_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="noise")
|
|
snap_color_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="snap_color")
|
|
one_bit_data = filtered(perturbed_data_normalized, len(perturbed_data_normalized), filter="1-bit")
|
|
|
|
# evaluate the model on the attacked and filtered images
|
|
output_unfiltered = model(perturbed_data_normalized)
|
|
output_kuwahara = model(kuwahara_data)
|
|
output_bilateral = model(bilateral_data)
|
|
output_gaussian_blur = model(gaussian_blur_data)
|
|
output_random_noise = model(random_noise_data)
|
|
output_snap_color = model(snap_color_data)
|
|
output_one_bit = model(one_bit_data)
|
|
|
|
# Get the predicted class from the model for each case
|
|
unfiltered_pred = output_unfiltered.max(1, keepdim=True)[1]
|
|
kuwahara_pred = output_kuwahara.max(1, keepdim=True)[1]
|
|
bilateral_pred = output_bilateral.max(1, keepdim=True)[1]
|
|
gaussian_blur_pred = output_gaussian_blur.max(1, keepdim=True)[1]
|
|
random_noise_pred = output_random_noise.max(1, keepdim=True)[1]
|
|
snap_color_pred = output_snap_color.max(1, keepdim=True)[1]
|
|
one_bit_pred = output_one_bit.max(1, keepdim=True)[1]
|
|
predictions = [unfiltered_pred.item(), kuwahara_pred.item(), bilateral_pred.item(), gaussian_blur_pred.item(), random_noise_pred.item(), snap_color_pred.item(), one_bit_pred.item()]
|
|
plurality_pred = stats.mode(predictions)[0]
|
|
|
|
# Count up correct classifications for each case
|
|
if orig_pred.item() == target.item():
|
|
orig_correct += 1
|
|
|
|
if unfiltered_pred.item() == target.item():
|
|
unfiltered_correct += 1
|
|
|
|
if kuwahara_pred.item() == target.item():
|
|
kuwahara_correct += 1
|
|
|
|
if bilateral_pred.item() == target.item():
|
|
bilateral_correct += 1
|
|
|
|
if gaussian_blur_pred.item() == target.item():
|
|
gaussian_blur_correct += 1
|
|
|
|
if random_noise_pred.item() == target.item():
|
|
random_noise_correct += 1
|
|
|
|
if snap_color_pred.item() == target.item():
|
|
snap_color_correct += 1
|
|
|
|
if one_bit_pred.item() == target.item():
|
|
one_bit_correct += 1
|
|
|
|
if plurality_pred == target.item():
|
|
plurality_correct += 1
|
|
|
|
# Calculate the overall accuracy of each case
|
|
orig_acc = orig_correct/float(len(test_loader))
|
|
unfiltered_acc = unfiltered_correct/float(len(test_loader))
|
|
kuwahara_acc = kuwahara_correct/float(len(test_loader))
|
|
bilateral_acc = bilateral_correct/float(len(test_loader))
|
|
gaussian_blur_acc = gaussian_blur_correct/float(len(test_loader))
|
|
random_noise_acc = random_noise_correct/float(len(test_loader))
|
|
snap_color_acc = snap_color_correct/float(len(test_loader))
|
|
one_bit_acc = one_bit_correct/float(len(test_loader))
|
|
plurality_acc = plurality_correct/float(len(test_loader))
|
|
|
|
print(f"====== EPSILON: {epsilon} ======")
|
|
print(f"Clean (No Filter) Accuracy = {orig_correct} / {len(test_loader)} = {orig_acc}")
|
|
print(f"Unfiltered Accuracy = {unfiltered_correct} / {len(test_loader)} = {unfiltered_acc}")
|
|
print(f"Kuwahara Filter Accuracy = {kuwahara_correct} / {len(test_loader)} = {kuwahara_acc}")
|
|
print(f"Bilateral Filter Accuracy = {bilateral_correct} / {len(test_loader)} = {bilateral_acc}")
|
|
print(f"Gaussian Blur Accuracy = {gaussian_blur_correct} / {len(test_loader)} = {gaussian_blur_acc}")
|
|
print(f"Random Noise Accuracy = {random_noise_correct} / {len(test_loader)} = {random_noise_acc}")
|
|
print(f"Snapped Color Accuracy = {snap_color_correct} / {len(test_loader)} = {snap_color_acc}")
|
|
print(f"1 Bit Accuracy = {one_bit_correct} / {len(test_loader)} = {one_bit_acc}")
|
|
print(f"Plurality Vote Accuracy = {plurality_correct} / {len(test_loader)} = {plurality_acc}")
|
|
|
|
return unfiltered_acc, kuwahara_acc, bilateral_acc, gaussian_blur_acc, random_noise_acc, snap_color_acc, one_bit_acc, plurality_acc
|
|
|
|
|
|
def filtered(data, batch_size=64, filter="kuwahara"):
|
|
# Turn the tensor into an image
|
|
images = None
|
|
try:
|
|
images = data.numpy().transpose(0,2,3,1)
|
|
except RuntimeError:
|
|
images = data.detach().numpy().transpose(0,2,3,1)
|
|
|
|
|
|
# Apply the Kuwahara filter
|
|
filtered_images = np.ndarray((batch_size,28,28,1))
|
|
|
|
if filter == "kuwahara":
|
|
for i in range(batch_size):
|
|
filtered_images[i] = kuwahara(images[i], method='gaussian', radius=5, image_2d=images[i])
|
|
elif filter == "aniso_diff":
|
|
for i in range(batch_size):
|
|
img_3ch = np.zeros((np.array(images[i]), np.array(images[i]).shape[1], 3))
|
|
img_3ch[:,:,0] = images[i]
|
|
img_3ch[:,:,1] = images[i]
|
|
img_3ch[:,:,2] = images[i]
|
|
img_3ch_filtered = cv2.ximgproc.anisotropicDiffusion(img2, alpha=0.2, K=0.5, niters=5)
|
|
filtered_images[i] = cv2.cvtColor(img_3ch_filtered, cv2.COLOR_RGB2GRAY)
|
|
plt.imshow(filtered_images[i])
|
|
plt.show()
|
|
elif filter == "noise":
|
|
for i in range(batch_size):
|
|
mean = 0
|
|
stddev = 180
|
|
noise = np.zeros(images[i].shape, images[i].dtype)
|
|
cv2.randn(noise, mean, stddev)
|
|
filtered_images[i] = cv2.addWeighted(images[i], 1.0, noise, 0.001, 0.0).reshape(filtered_images[i].shape)
|
|
elif filter == "gaussian_blur":
|
|
for i in range(batch_size):
|
|
filtered_images[i] = cv2.GaussianBlur(images[i], ksize=(5,5), sigmaX=0).reshape(filtered_images[i].shape)
|
|
elif filter == "bilateral":
|
|
for i in range(batch_size):
|
|
filtered_images[i] = cv2.bilateralFilter(images[i], 5, 50, 50).reshape(filtered_images[i].shape)
|
|
elif filter == "1-bit":
|
|
num_colors = 2
|
|
for i in range(batch_size):
|
|
# If the channel contains any negative values, define the lowest negative value as black
|
|
min_value = np.min(images[i])
|
|
if (min_value < 0):
|
|
filtered_images[i] = images[i] + min_value
|
|
|
|
# If the color space extends beyond [0,1], re-scale all of the colors to that range
|
|
max_value = np.max(filtered_images[i])
|
|
if (max_value > 1):
|
|
filtered_images[i] *= (num_colors/max_value)
|
|
filtered_images[i] = filtered_images[i].astype(int).astype(float)*(max_value/num_colors)
|
|
else:
|
|
filtered_images[i] *= num_colors
|
|
filtered_images[i] = filtered_images[i].astype(int).astype(float)/num_colors
|
|
if (min_value < 0):
|
|
filtered_images[i] -= min_value
|
|
elif filter == "snap_color":
|
|
for i in range(batch_size):
|
|
filtered_images[i] = (images[i]*4).astype(int).astype(float) / 4
|
|
|
|
# Modify the data with the filtered image
|
|
filtered_images = filtered_images.transpose(0,3,1,2)
|
|
return torch.tensor(filtered_images).float()
|
|
|
|
unfiltered_accuracies = []
|
|
kuwahara_accuracies = []
|
|
bilateral_accuracies = []
|
|
gaussian_blur_accuracies = []
|
|
random_noise_accuracies = []
|
|
snap_color_accuracies = []
|
|
one_bit_accuracies = []
|
|
pluality_vote_accuracies = []
|
|
|
|
print(f"Model: {pretrained_model}")
|
|
|
|
for eps in epsilons:
|
|
unfiltered_acc, kuwahara_acc, bilateral_acc, gaussian_blur_acc, random_noise_acc, snap_color_acc, one_bit_acc, plurality_acc = test(model, device, test_loader, eps)
|
|
unfiltered_accuracies.append(unfiltered_acc)
|
|
kuwahara_accuracies.append(kuwahara_acc)
|
|
bilateral_accuracies.append(bilateral_acc)
|
|
gaussian_blur_accuracies.append(gaussian_blur_acc)
|
|
random_noise_accuracies.append(random_noise_acc)
|
|
snap_color_accuracies.append(snap_color_acc)
|
|
one_bit_accuracies.append(one_bit_acc)
|
|
pluality_vote_accuracies.append(plurality_acc)
|
|
|
|
# Plot the results
|
|
plt.plot(epsilons, unfiltered_accuracies, label="Attacked Accuracy")
|
|
plt.plot(epsilons, kuwahara_accuracies, label="Kuwahara Accuracy")
|
|
plt.plot(epsilons, bilateral_accuracies, label="Bilateral Accuracy")
|
|
plt.plot(epsilons, gaussian_blur_accuracies, label="Gaussian Blur Accuracy")
|
|
plt.plot(epsilons, random_noise_accuracies, label="Random Noise Accuracy")
|
|
plt.plot(epsilons, snap_color_accuracies, label="Snapped Color Accuracy")
|
|
plt.plot(epsilons, one_bit_accuracies, label="1-Bit Accuracy")
|
|
plt.plot(epsilons, pluality_vote_accuracies, label="Plurality Vote Accuracy")
|
|
plt.legend()
|
|
|
|
plt.show()
|
|
|