-
Notifications
You must be signed in to change notification settings - Fork 70
/
Copy pathEIT.py
188 lines (147 loc) · 8.58 KB
/
EIT.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Reference: C. Guo et al., "Countering Adversarial Images Using Input Transformations" in ICLR, 2018.
# Reference Implementation from Authors (TensorFlow): https://github.com/facebookarchive/adversarial_image_defenses
# **************************************
# @Time : 2018/11/23 20:30
# @Author : Jiaxu Zou
# @Lab : nesa.zju.edu.cn
# @File : EIT.py
# **************************************
import os
from tqdm import tqdm
import numpy as np
import torch
import torch.optim as optim
from PIL import Image
from torchvision.transforms import Compose, RandomAffine, RandomCrop, RandomHorizontalFlip, Resize, ToPILImage, ToTensor
from Defenses.DefenseMethods.defenses import Defense
from RawModels.ResNet import adjust_learning_rate
from RawModels.Utils.TrainTest import train_one_epoch, validation_evaluation
def image_crop_rescale(sample, crop_size, color_mode):
image = ToPILImage(mode=color_mode)(sample)
cropped_image = RandomCrop(crop_size)(image)
rescaled_image = Resize((sample.shape[1], sample.shape[2]), interpolation=0)(cropped_image)
cropped_rescaled_sample = ToTensor()(rescaled_image)
return cropped_rescaled_sample
def bit_depth_reduction(samples, depth):
level = 2 ** depth
reduced_images = np.rint(samples * (level - 1)) / (level - 1)
return reduced_images
def total_variance_minimization(image, lambda_tv):
from Defenses.DefenseMethods.External.InputTransformations import defend_tv
image_numpy_channel = image.permute(1, 2, 0).numpy()
tv_min_image = defend_tv(input_array=image_numpy_channel, lambda_tv=lambda_tv)
return torch.from_numpy(tv_min_image).float().permute(2, 0, 1)
def jpeg_compress(image, quality, color_mode):
from Defenses.DefenseMethods.External.InputTransformations import defend_jpeg
return defend_jpeg(input_tensor=image, image_mode=color_mode, quality=quality)
# For help when requiring self-defined dataset provision with batches during training
class TransformedDataset(torch.utils.data.Dataset):
def __init__(self, images, labels, dataset, transform):
super(TransformedDataset, self).__init__()
self.images = images
self.labels = labels
self.transform = transform
self.color_mode = 'RGB' if dataset == 'CIFAR10' else 'L'
def __getitem__(self, index):
single_image, single_label = self.images[index], self.labels[index]
if self.transform:
img = ToPILImage(mode=self.color_mode)(single_image)
single_image = self.transform(img)
return single_image, single_label
def __len__(self):
return len(self.images)
class EITDefense(Defense):
def __init__(self, model=None, defense_name=None, dataset=None, re_training=True, training_parameters=None, device=None, **kwargs):
super(EITDefense, self).__init__(model=model, defense_name=defense_name)
self.model = model
self.defense_name = defense_name
self.device = device
self.Dataset = dataset.upper()
assert self.Dataset in ['MNIST', 'CIFAR10'], "The data set must be MNIST or CIFAR10"
# make sure to parse the parameters for the defense
assert self._parsing_parameters(**kwargs)
if re_training:
# get the training_parameters, the same as the settings of RawModels
self.num_epochs = training_parameters['num_epochs']
self.batch_size = training_parameters['batch_size']
# prepare the optimizers
if self.Dataset == 'MNIST':
self.optimizer = optim.SGD(self.model.parameters(), lr=training_parameters['learning_rate'],
momentum=training_parameters['momentum'], weight_decay=training_parameters['decay'], nesterov=True)
else:
self.optimizer = optim.Adam(self.model.parameters(), lr=training_parameters['lr'])
self.color_mode = 'RGB' if self.Dataset == 'CIFAR10' else 'L'
if self.Dataset == 'CIFAR10':
self.transform = Compose([
RandomAffine(degrees=0, translate=(0.1, 0.1)),
RandomHorizontalFlip(),
ToTensor()
])
else:
self.transform = None
def _parsing_parameters(self, **kwargs):
assert kwargs is not None, "the parameters should be specified"
print("\nUser configurations for the {} defense".format(self.defense_name))
for key in kwargs:
print('\t{} = {}'.format(key, kwargs[key]))
self.bit_depth = kwargs['bit_depth']
self.crop_size = kwargs['crop_size']
self.JPEG_quality = kwargs['JPEG_quality']
self.lambda_tv = kwargs['lambda_tv']
return True
def ensemble_input_transformations(self, images):
transformed_batch_images_list = []
for index in range(images.shape[0]):
image = images[index]
image = torch.from_numpy(image).to('cpu')
# Image Crop and Rescaling
cropped_rescaled_image = image_crop_rescale(sample=image, crop_size=self.crop_size, color_mode=self.color_mode)
# Total Variance Minimization
tv_mim_image = total_variance_minimization(image=cropped_rescaled_image, lambda_tv=self.lambda_tv)
# JPEG Compression
compressed_image = jpeg_compress(image=tv_mim_image, quality=self.JPEG_quality, color_mode=self.color_mode)
transformed_batch_images_list.append(compressed_image.numpy())
transformed_batch_images_numpy = np.array(transformed_batch_images_list)
# bit depth for batch images
transformed_batch_images_numpy = bit_depth_reduction(transformed_batch_images_numpy, depth=self.bit_depth)
return transformed_batch_images_numpy
def transforming_dataset(self, data_loader=None):
transformed_data = []
transformed_label = []
print('\ntransforming dataset ....\n')
for index, (images, labels) in enumerate(tqdm(data_loader)):
np_images = images.cpu().numpy()
np_labels = labels.cpu().numpy()
transformed_image_numpy = self.ensemble_input_transformations(images=np_images)
transformed_data.extend(transformed_image_numpy)
transformed_label.extend(np_labels)
return np.array(transformed_data), np.array(transformed_label)
def defense(self, train_loader=None, valid_loader=None):
transformed_train_data_numpy, transformed_train_label_numpy = self.transforming_dataset(train_loader)
transformed_val_data_numpy, transformed_val_label_numpy = self.transforming_dataset(valid_loader)
transformed_train_dataset = TransformedDataset(images=torch.from_numpy(transformed_train_data_numpy),
labels=torch.from_numpy(transformed_train_label_numpy), dataset=self.Dataset,
transform=self.transform)
transformed_train_loader = torch.utils.data.DataLoader(transformed_train_dataset, batch_size=self.batch_size, shuffle=True)
transformed_val_dataset = TransformedDataset(images=torch.from_numpy(transformed_val_data_numpy),
labels=torch.from_numpy(transformed_val_label_numpy), dataset=self.Dataset,
transform=None)
transformed_val_loader = torch.utils.data.DataLoader(transformed_val_dataset, batch_size=self.batch_size, shuffle=False)
best_val_acc = None
for epoch in range(self.num_epochs):
train_one_epoch(model=self.model, train_loader=transformed_train_loader, optimizer=self.optimizer, epoch=epoch, device=self.device)
val_acc = validation_evaluation(model=self.model, validation_loader=transformed_val_loader, device=self.device)
if self.Dataset == 'CIFAR10':
adjust_learning_rate(epoch=epoch, optimizer=self.optimizer)
# save the retrained defense-enhanced model
assert os.path.exists('../DefenseEnhancedModels/{}'.format(self.defense_name))
defense_enhanced_saver = '../DefenseEnhancedModels/{}/{}_{}_enhanced.pt'.format(self.defense_name, self.Dataset, self.defense_name)
if not best_val_acc or round(val_acc, 4) >= round(best_val_acc, 4):
if best_val_acc is not None:
os.remove(defense_enhanced_saver)
best_val_acc = val_acc
self.model.save(name=defense_enhanced_saver)
else:
print('Train Epoch{:>3}: validation dataset accuracy did not improve from {:.4f}\n'.format(epoch, best_val_acc))