/
trainPoison.py
167 lines (132 loc) · 6.52 KB
/
trainPoison.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
import argparse
import itertools
import os
import pdb
import sys
from pprint import pprint
import torch
import torch.nn as nn
import torch.optim as optim
import torch.optim.lr_scheduler as scheduler
import torchvision.transforms as transforms
import torchvision.utils
from torch.utils.data import DataLoader
from tqdm import tqdm
from PIL import Image
from Models import *
import dataLoader
import Utilities
def evaluateModel(args, model, device, trainData, validationData, target, targetClass):
trainAccuracy = Utilities.classificationAccuracy(model, device, trainData)
print("Train Accuracy: " + str(trainAccuracy))
Utilities.writeLog(args.logFileLocation, "Train Accuracy: " + str(trainAccuracy))
testAccuracy = Utilities.classificationAccuracy(model, device, validationData)
print("Test Accuracy: " + str(testAccuracy))
Utilities.writeLog(args.logFileLocation, "Test Accuracy: " + str(testAccuracy))
targetSuccess = Utilities.testTarget(model, device, target, targetClass)
print("Poison Success on Target Image: " + str(targetSuccess))
Utilities.writeLog(args.logFileLocation, "Poison Success on Target Image: " + str(targetSuccess))
def train(args):
modelName = args.architecture
checkPointDirectory = args.checkPointDirectory
dataSplitDirectory = args.dataSplitDirectory
poisonIndex = args.poisonIndex
CUDA = args.CUDA
BATCHSIZE = args.batchSize
EPOCH = args.numEpoch
WORKERS = args.workers
LR = args.learningRate
WEIGHTDECAY = args.weightDecay
IMGSIZE = args.imageSize
targetClass = 8 #Ships
print("Using Parameters:")
pprint(vars(args))
START = 0
normalize = transforms.Normalize(mean=[0.4914, 0.4822, 0.4465], std=[0.2023, 0.1994, 0.2010])
dataAugmentation = transforms.Compose([transforms.Resize((IMGSIZE, IMGSIZE)), transforms.ToTensor(), normalize])
try:
Utilities.writeLog(args.logFileLocation, "Model Architecture: " + str(modelName))
Utilities.writeLog(args.logFileLocation, "Poison Index: " + str(poisonIndex))
Utilities.writeLog(args.logFileLocation, "K Value: " + str(args.K))
Utilities.writeLog(args.logFileLocation, "classBalance: " + str(args.classBalance))
Utilities.writeLog(args.logFileLocation, "Replicate Imbalance: " + str(args.replicateImbalance))
print("Poison Index: " + str(poisonIndex))
TrainDataset = dataLoader.ConvexPolytopeFineTune_DataLoader(args, poisonIndex, dataAugmentation)
trainData = DataLoader(TrainDataset, batch_size=BATCHSIZE, shuffle=True, num_workers=WORKERS)
ValidationDataset = dataLoader.ConvexPolytopeEvaluation_DataLoader(dataSplitDirectory + "testFile.txt", dataAugmentation)
validationData = DataLoader(ValidationDataset, batch_size=BATCHSIZE, shuffle=True, num_workers=WORKERS)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if modelName == "DenseNet121":
model = DenseNet121()
elif modelName == "DPN92":
model = DPN92()
elif modelName == "GoogLeNet":
model = GoogLeNet()
elif modelName == "MobileNetV2":
model = MobileNetV2()
elif modelName == "ResNet18":
model = ResNet18()
elif modelName == "ResNet50":
model = ResNet50()
elif modelName == "ResNeXt29_2x64d":
model = ResNeXt29_2x64d()
elif modelName == "SENet18":
model = SENet18()
trainableParameters = model.get_penultimate_params_list()
optimizer = optim.Adam(trainableParameters, lr=LR, weight_decay=WEIGHTDECAY)
CEL = nn.CrossEntropyLoss()
if CUDA and torch.cuda.device_count() > 1:
model = nn.DataParallel(model)
CheckPoint = torch.load(checkPointDirectory + "/" + modelName + ".pth")
model.load_state_dict(CheckPoint["net"])
model.to(device)
if CUDA and torch.cuda.device_count() > 1:
trainableParameters = model.module.get_penultimate_params_list()
optimizer = optim.Adam(trainableParameters, lr=LR, weight_decay=WEIGHTDECAY)
model.to(device)
for STEP in range(START, EPOCH):
epochLoss = 0
model.train()
for batchCount, data in enumerate(trainData, 1):
img = data[0].to(device)
classID = data[1].to(device)
classification = model(img)
optimizer.zero_grad()
Loss = CEL(classification, classID)
epochLoss = epochLoss + Loss.item()
Loss.backward()
optimizer.step()
print("Epoch: " + str(STEP + 1) + " of " + str(EPOCH) + " | " + "Epoch Loss: " + str(epochLoss / batchCount))
targetFile = Utilities.parseTargetIndex(dataSplitDirectory + "targetImageFile.txt", poisonIndex)
target = dataAugmentation(Image.open(targetFile))
evaluateModel(args, model, device, trainData, validationData, target, targetClass)
print("Status: Experiment Completed Successfully")
Utilities.writeLog(args.logFileLocation, "Status: Experiment Completed Successfully")
except Exception as e:
print("Status: " + str(e))
Utilities.writeLog(args.logFileLocation, "Status: " + str(e))
print("\n")
parser = argparse.ArgumentParser()
parser.add_argument("--logFileLocation", required=True)
parser.add_argument("--architecture", required=True)
parser.add_argument("--checkPointDirectory", required=True)
parser.add_argument("--poisonIndex", required=True, type=int)
parser.add_argument("--poisonImageDirectory", required=True)
parser.add_argument("--dataSplitDirectory", required=True)
parser.add_argument("--featureDirectory", required=True)
parser.add_argument("--CUDA", dest="CUDA", action="store_true")
parser.add_argument("--CPU", dest="CUDA", action="store_false")
parser.add_argument("--replicateImbalance", dest="replicateImbalance", action="store_true")
parser.add_argument("--classBalance", default=[50, 50, 50, 50, 50, 50, 50, 50, 50, 50], nargs="+", type=int)
parser.add_argument("--batchSize", default=64, type=int)
parser.add_argument("--numEpoch", default=60, type=int)
parser.add_argument("--workers", default=0, type=int)
parser.add_argument("--learningRate", default=0.1, type=float)
parser.add_argument("--K", default=101, type=int)
parser.add_argument("--weightDecay", default=0, type=float)
parser.add_argument("--imageSize", default=32, type=int)
parser.set_defaults(CUDA=True)
parser.set_defaults(replicateImbalance=False)
args = parser.parse_args()
if __name__ == "__main__":
train(args)