DeepfoolSingleImageTest

import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.optim as optim
import torch.utils.data as data_utils
from torch.autograd import Variable
import math
import torchvision.models as models
from PIL import Image
from deepfool import deepfool
import os
import time


net = models.resnet34(pretrained=True)
# Switch to evaluation mode
net.eval()

im_orig = Image.open('pictures/test_im2.jpg')

mean = [ 0.485, 0.456, 0.406 ]
std = [ 0.229, 0.224, 0.225 ]

im2 = transforms.Compose([
    transforms.Scale(256),
    transforms.CenterCrop(224)])(im_orig)

# Remove the mean
im = transforms.Compose([
    transforms.Scale(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean = mean,
                         std = std)])(im_orig)

#start time
start_time = time.time()
r, loop_i, label_orig, label_pert, pert_image, pert = deepfool(im, net)
end_time = time.time()
execution_time = end_time - start_time
print("execution time = " + str(execution_time))
#end time
#execution time = end time - start time

labels = open(os.path.join('synset_words.txt'), 'r').read().split('\n')

str_label_orig = labels[np.int(label_orig)].split(',')[0]
str_label_pert = labels[np.int(label_pert)].split(',')[0]

print("Original label = ", str_label_orig)
print("Perturbed label = ", str_label_pert)

def clip_tensor(A, minv, maxv):
    A = torch.max(A, minv*torch.ones(A.shape))
    A = torch.min(A, maxv*torch.ones(A.shape))
    return A

clip = lambda x: clip_tensor(x, 0, 255)

tf = transforms.Compose([transforms.Normalize(mean=[0, 0, 0], std=list(map(lambda x: 1 / x, std))),
                        transforms.Normalize(list(map(lambda x: -x, mean)), std=[1, 1, 1]),
                        transforms.Lambda(clip),
                        transforms.ToPILImage(),
                        transforms.CenterCrop(224)])

imagetransform = transforms.Compose([transforms.Normalize(mean=[0, 0, 0], std=list(map(lambda x: 1 / x, std))),
                        transforms.Normalize(list(map(lambda x: -x, mean)), std=[1, 1, 1]),
                        transforms.Lambda(clip)])

tensortransform = transforms.Compose([transforms.Scale(256),
                        transforms.CenterCrop(224),
                        transforms.ToTensor(),
                        transforms.Normalize(mean=[0, 0, 0], std=list(map(lambda x: 1 / x, std))),
                        transforms.Normalize(list(map(lambda x: -x, mean)), std=[1, 1, 1]),
                        transforms.Lambda(clip)])


def inv_tf_pert(r):

    pert = np.sum(np.absolute(r), axis=0)
    pert[pert != 0] = 1

    return pert


plt.figure()
plt.imshow(im2)
plt.title(str_label_orig)
plt.show()

plt.figure()
plt.imshow(tf(pert_image.cpu()[0]))
tf(pert_image.cpu()[0]).save('dfimg.png')
plt.title(str_label_pert)
plt.show()

plt.figure()
plt.imshow(tf(pert.cpu()[0]*1))
plt.title(str_label_orig)
plt.show()
tf(pert.cpu()[0]*1).save('dfpert.png')