Directory structure overhaul, poster almost done

.~lock.Poster.pptx#

@@ -1 +0,0 @@
-,sharpe,dhcp-150-250-208-169,30.04.2024 21:44,file:///home/sharpe/.var/app/org.libreoffice.LibreOffice/config/libreoffice/4;

src/defense_filters.py

@@ -1,3 +1,4 @@
+import matplotlib.pyplot as plt
 import cv2
 from pykuwahara import kuwahara
 import numpy as np
@@ -26,6 +27,10 @@ def gaussian_kuwahara(data, dimensions, radius=5):
 
     for i in range(dimensions[0]):
         filtered_images[i] = kuwahara(images[i], method='gaussian', radius=radius, image_2d=images[i])
+        if i == 0 and radius == 5:
+            plt.title("Gaussian Kuwahara", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
     
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
@@ -37,6 +42,10 @@ def mean_kuwahara(data, dimensions, radius=5):
 
     for i in range(dimensions[0]):
         filtered_images[i] = kuwahara(images[i], method='mean', radius=radius, image_2d=images[i])
+        if i == 0 and radius == 5:
+            plt.title("Mean Kuwahara", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
     
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
@@ -52,6 +61,11 @@ def random_noise(data, dimensions, intensity=0.001):
         noise = np.zeros(images[i].shape, images[i].dtype)
         cv2.randn(noise, mean, stddev)
         filtered_images[i] = cv2.addWeighted(images[i], 1.0, noise, intensity, 0.0).reshape(filtered_images[i].shape)
+        if i == 0 and intensity == 0.0015:
+            plt.title("Random Noise", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
+
 
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
@@ -63,6 +77,13 @@ def gaussian_blur(data, dimensions, ksize=(5,5)):
 
     for i in range(dimensions[0]):
         filtered_images[i] = cv2.GaussianBlur(images[i], ksize=ksize, sigmaX=0).reshape(filtered_images[i].shape)
+        if i == 0 and ksize[0] == 5:
+            plt.title("Unfiltered", fontsize=40)
+            plt.imshow(images[i], cmap='gray')
+            plt.show()
+            plt.title("Gaussian Blur", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
 
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
@@ -74,6 +95,10 @@ def bilateral_filter(data, dimensions, d=5, sigma=50):
 
     for i in range(dimensions[0]):
         filtered_images[i] = cv2.bilateralFilter(images[i], d, sigma, sigma).reshape(filtered_images[i].shape)
+        if i == 0 and d == 5:
+            plt.title("Bilateral Filter", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
 
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
@@ -100,6 +125,11 @@ def threshold_filter(data, dimensions, threshold=0.5):
         if min_value < 0:
             filtered_images[i] -= min_value
 
+        if i == 0 and threshold == 0.5:
+            plt.title("Threshold Filter", fontsize=40)
+            plt.imshow(filtered_images[i], cmap='gray')
+            plt.show()
+    
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
 
@@ -111,6 +141,12 @@ def bit_depth(data, dimensions, bits=16):
     for i in range(dimensions[0]):
         filtered_images[i] = (images[i]*(2**bits)).astype(int).astype(float)/(2**bits)
 
+
+    if i == 0 and bits == 4:
+        plt.title("Bit-Depth", fontsize=40)
+        plt.imshow(filtered_images[i], cmap='gray')
+        plt.show()
+
     filtered_images = filtered_images.transpose(0,3,1,2)
     return torch.tensor(filtered_images).float()
 

src/test_defenses.py

@@ -20,39 +20,40 @@ import defense_filters
 
 
 ATTACK = "FGSM"
-DATASET = "CIFAR-10"
+DATASET = "MNIST"
 
-RES_X = 32
-RES_Y = 32
-CHANNELS = 3
+RES_X = 28
+RES_Y = 28
+CHANNELS = 1
 
+MIN_EPSILON = 0.2
 MAX_EPSILON = 0.3
 EPSILON_STEP = 0.025
 
 TESTED_STRENGTH_COUNT = 5
-epsilons = np.arange(0.0, MAX_EPSILON+EPSILON_STEP, EPSILON_STEP)
-pretrained_model = "cifar_vgg.pth"
+epsilons = np.arange(MIN_EPSILON, MAX_EPSILON+EPSILON_STEP, EPSILON_STEP)
+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)
+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=False)
 
-transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
-batch_size = 1
-testset = datasets.CIFAR10(root='./data', train=False, download=True, transform=transform)
-test_loader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=True, num_workers=2)
+#transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
+#batch_size = 1
+#testset = datasets.CIFAR10(root='./data', train=False, download=True, transform=transform)
+#test_loader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=True, num_workers=2)
 
 print("CUDA Available: ", torch.cuda.is_available())
 device = torch.device("cuda" if use_cuda and torch.cuda.is_available() else "cpu")
 
-model = vgg.VGG("VGG16").to(device)
+model = mnist.Net().to(device)
 
 model.load_state_dict(torch.load(pretrained_model, map_location=device))
 
@@ -150,6 +151,7 @@ def test(model, device, test_loader, epsilon, filter):
                 while i >= len(filtered_correct_counts):
                     filtered_correct_counts.append(0)
                 filtered_correct_counts[i] += 1
+        return
 
         # Get the predicted classification
         unfiltered_pred = output_unfiltered.max(1, keepdim=True)[1]
@@ -199,8 +201,9 @@ for filter in filters:
         results["filters"][filter].append(accuracies)
 
         results_json = json.dumps(results, indent=4)
-        with open("results/cifar10_fgsm.json", "w") as outfile:
-            outfile.write(results_json)
+        break
+        #with open("results/cifar10_fgsm.json", "w") as outfile:
+        #    outfile.write(results_json)
 
 # Plot the results
 #plt.figure(figsize=(16,9))