/
verify.py
254 lines (222 loc) · 14.6 KB
/
verify.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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
from __future__ import print_function
import argparse
import torch
import models
import os
import losses
import data_loader
from torchvision import transforms
import numpy
import models
import torchmetrics
import timm
numpy.set_printoptions(edgeitems=5, linewidth=160, formatter={'float': '{:0.6f}'.format})
torch.set_printoptions(edgeitems=5, precision=6, linewidth=160)
parser = argparse.ArgumentParser(description='OOD Performance Estimation Verifier')
parser.add_argument('-bs', '--batch_size', type=int, default=64, metavar='N', help='batch size for data loader')
parser.add_argument('--dataset', required=True, help='cifar10 | cifar100 | imagenet1k')
parser.add_argument('--dataroot', default='data', help='path to dataset')
parser.add_argument('--net_type', required=True, help='resnet | wideresnet')
parser.add_argument('--gpu', type=int, default=0, help='gpu index')
parser.add_argument('--loss', required=True, help='the loss used')
parser.add_argument('--dir', default="", type=str, help='Part of the dir to use')
parser.add_argument('-x', '--executions', default=1, type=int, metavar='N', help='Number of executions (default: 1)')
args = parser.parse_args()
torch.cuda.set_device(args.gpu)
def main():
print("\n\n\n\n\n")
print("#########################")
print("#########################")
print("######## VERIFY #########")
print("#########################")
print("#########################")
print(args)
dir_path = os.path.join("experiments", args.dir, "train_classify", "data~"+args.dataset+"+model~"+args.net_type+"+loss~"+str(args.loss))
file_path = os.path.join(dir_path, "results_threshold.csv")
with open(file_path, "w") as results_file:
results_file.write("EXECUTION,MODEL,IN-DATA,OUT-DATA,LOSS,SCORE,PERCENTILE,FPR,TPR,THRESHOLD\n")
file_path2 = os.path.join(dir_path, "results_auroc.csv")
with open(file_path2, "w") as results_file:
results_file.write("EXECUTION,MODEL,IN-DATA,OUT-DATA,LOSS,SCORE,AUROC\n")
# define number of classes
if args.dataset == 'cifar100':
args.num_classes = 100
elif args.dataset == 'imagenet1k':
args.num_classes = 1000
else:
args.num_classes = 10
if args.dataset == 'cifar10':
out_dist_list = ['cifar100', 'imagenet_resize', 'lsun_resize', 'svhn']
elif args.dataset == 'cifar100':
out_dist_list = ['cifar10', 'imagenet_resize', 'lsun_resize', 'svhn']
elif args.dataset == 'imagenet1k':
out_dist_list = ['imagenet-o']
if args.dataset == 'cifar10':
in_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.491, 0.482, 0.446), (0.247, 0.243, 0.261))])
elif args.dataset == 'cifar100':
in_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.507, 0.486, 0.440), (0.267, 0.256, 0.276))])
elif args.dataset == 'imagenet1k':
args.dataroot = '/mnt/ssd/imagenet1k'
args.input_size = 224
args.DEFAULT_CROP_RATIO = 0.875
in_transform = transforms.Compose([
transforms.Resize(int(args.input_size / args.DEFAULT_CROP_RATIO)),
transforms.CenterCrop(args.input_size),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])
for args.execution in range(1, args.executions + 1):
print("\nEXECUTION:", args.execution)
pre_trained_net = os.path.join(dir_path, "model" + str(args.execution) + ".pth")
loss_first_part = losses.DisMaxLossFirstPart
scores = ["MMLES"]
# load networks
if args.net_type == 'resnet34':
model = models.ResNet34(num_c=args.num_classes, loss_first_part=loss_first_part)
elif args.net_type == 'densenetbc100':
model = models.DenseNet3(100, int(args.num_classes), loss_first_part=loss_first_part)
elif args.net_type == "wideresnet2810":
model = models.Wide_ResNet(depth=28, widen_factor=10, num_classes=args.num_classes, loss_first_part=loss_first_part)
elif args.net_type == "resnet18":
model = timm.create_model('resnet18', pretrained=False)
num_in_features = model.get_classifier().in_features
model.fc = loss_first_part(num_in_features, args.num_classes)
model.load_state_dict(torch.load(pre_trained_net, map_location="cuda:" + str(args.gpu)))
model.cuda()
print('load model: ' + args.net_type)
model.eval()
print(model.classifier.validationset_available.item())
# load dataset
print('load target valid data: ', args.dataset)
_, test_loader = data_loader.getTargetDataSet(args, args.dataset, args.batch_size, in_transform, args.dataroot)
for score in scores:
print("###############################")
print("###############################")
print("SCORE:", score)
print("###############################")
print("###############################")
i = 1 if model.classifier.validationset_available.item() else 0
print("Precomputed Thresholds (initial):\n", model.classifier.precomputed_thresholds[i].detach().cpu().numpy())
for out_dist in out_dist_list:
out_test_loader = data_loader.getNonTargetDataSet(args, out_dist, args.batch_size, in_transform, args.dataroot)
print('Out-distribution: ' + out_dist)
auroc, roc = detect(model, test_loader, out_test_loader, score)
print("AUROC: ", auroc.item())
fpr_list = []
tpr_list = []
threshold_list = []
for j in range(25):
index = find_index_of_nearest(model.classifier.precomputed_thresholds[i,j].detach().cpu().numpy(), roc[2].detach().cpu().numpy())
fpr_list.append(roc[0][index].item())
tpr_list.append(roc[1][index].item())
threshold_list.append(roc[2][index].item())
print("FPRs:\n", fpr_list)
print("TPRs:\n", tpr_list)
print("Precomputed Threshold (final):\n", threshold_list)
#Saving regular estimations (requires in-distribution data and out-distribution data):
with open(file_path, "a") as results_file:
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0", fpr_list[0], tpr_list[0], 0))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0.1", fpr_list[1], tpr_list[1], 1))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0.2", fpr_list[2], tpr_list[2], 2))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0.3", fpr_list[3], tpr_list[3], 3))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0.4", fpr_list[4], tpr_list[4], 4))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "0.5", fpr_list[5], tpr_list[5], 5))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "1", fpr_list[6], tpr_list[6], 6))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "2", fpr_list[7], tpr_list[7], 7))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "3", fpr_list[8], tpr_list[8], 8))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "4", fpr_list[9], tpr_list[9], 9))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "5", fpr_list[10], tpr_list[10], 10))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "10", fpr_list[11], tpr_list[11], 11))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "15", fpr_list[12], tpr_list[12], 12))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "20", fpr_list[13], tpr_list[13], 13))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "25", fpr_list[14], tpr_list[14], 14))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "30", fpr_list[15], tpr_list[15], 15))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "35", fpr_list[16], tpr_list[16], 16))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "40", fpr_list[17], tpr_list[17], 17))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "45", fpr_list[18], tpr_list[18], 18))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "50", fpr_list[19], tpr_list[19], 19))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "60", fpr_list[20], tpr_list[20], 20))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "70", fpr_list[21], tpr_list[21], 21))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "80", fpr_list[22], tpr_list[22], 22))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "90", fpr_list[23], tpr_list[23], 23))
results_file.write("{},{},{},{},{},{},{},{},{},{}\n".format(
str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, "100", fpr_list[24], tpr_list[24], 24))
with open(file_path2, "a") as results_file:
results_file.write("{},{},{},{},{},{},{}\n".format(str(args.execution), args.net_type, args.dataset, out_dist, str(args.loss), score, auroc))
def detect(model, inloader, oodloader, score_type):
auroc = torchmetrics.AUROC(pos_label=1)
roc = torchmetrics.ROC(pos_label=1)
model.eval()
with torch.no_grad():
for _, (inputs, targets) in enumerate(inloader):
inputs, targets = inputs.to("cuda"), targets.to("cuda")
targets.fill_(1)
logits = model(inputs)
probabilities = torch.nn.Softmax(dim=1)(logits)
if score_type == "MPS":
scores = probabilities.max(dim=1)[0]
elif score_type == "ES":
scores = (probabilities * torch.log(probabilities)).sum(dim=1)
elif score_type == "MDS":
scores = logits.max(dim=1)[0]
elif score_type == "MMLS":
scores = logits.max(dim=1)[0] + logits.mean(dim=1)
elif score_type == "MMLES":
scores = logits.max(dim=1)[0] + logits.mean(dim=1) + (probabilities * torch.log(probabilities)).sum(dim=1)
elif score_type == "MMLEPS":
scores = logits.max(dim=1)[0] + logits.mean(dim=1) + (probabilities * torch.log(probabilities)).sum(dim=1) + probabilities.max(dim=1)[0]
auroc.update(scores, targets)
roc.update(scores, targets)
for _, (inputs, targets) in enumerate(oodloader):
inputs, targets = inputs.to("cuda"), targets.to("cuda")
targets.fill_(0)
logits = model(inputs)
probabilities = torch.nn.Softmax(dim=1)(logits)
if score_type == "MPS":
scores = probabilities.max(dim=1)[0]
elif score_type == "ES":
scores = (probabilities * torch.log(probabilities)).sum(dim=1)
elif score_type == "MDS":
scores = logits.max(dim=1)[0]
elif score_type == "MMLS":
scores = logits.max(dim=1)[0] + logits.mean(dim=1)
elif score_type == "MMLES":
scores = logits.max(dim=1)[0] + logits.mean(dim=1) + (probabilities * torch.log(probabilities)).sum(dim=1)
elif score_type == "MMLEPS":
scores = logits.max(dim=1)[0] + logits.mean(dim=1) + (probabilities * torch.log(probabilities)).sum(dim=1) + probabilities.max(dim=1)[0]
auroc.update(scores, targets)
roc.update(scores, targets)
return auroc.compute(), roc.compute()
def find_index_of_nearest(value, array):
array = numpy.asarray(array)
idx = (numpy.abs(array - value)).argmin()
return idx
if __name__ == '__main__':
main()