-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathtest.py
160 lines (126 loc) · 5.93 KB
/
test.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
import torch
from torch.autograd import Variable
from argparse import ArgumentParser
import numpy as np
from sklearn.metrics import roc_auc_score, roc_curve, auc
from models.Unet import *
from utils.utils import *
from dataloader import *
parser = ArgumentParser()
parser.add_argument('--config', type=str, default='configs/config_test.yaml', help="training configuration")
def find_fpr(fpr, tpr, tpr_target):
min_dist = 10000
min_idx = 0
for i in range(len(tpr)):
if tpr[i] > 0.95 and np.abs(tpr[i] - tpr_target) < min_dist:
min_dist = np.abs(tpr[i] - tpr_target)
min_idx = i
print("fpr = {:.4f} for tpr = {:.4f}".format(fpr[min_idx], tpr[min_idx]))
return fpr[min_idx], tpr[min_idx]
def test(model, target_class, perm_list, perm_cost, test_dataloader):
label_score_max = []
label_score_min = []
label_score_avg = []
model.eval()
for ind, data in enumerate(test_dataloader):
print('{}/10000'.format(ind * test_dataloader.batch_size))
inputs, labels = data
target = inputs
target = Variable(target).cuda()
partitioned_img, base = split_tensor(inputs, tile_size=inputs.size(2) // 2, offset=inputs.size(2) // 2)
min_score = torch.zeros(inputs.size(0)).cuda() + 100000
avg_score = torch.zeros(inputs.size(0)).cuda()
max_score = torch.zeros(inputs.size(0)).cuda()
idx = 0
num_perm = len(perm_list)
for perm in perm_list:
extended_perm = torch.tensor(perm) * inputs.size(1)
if inputs.size(1) == 3:
offset = torch.tensor([[0, 1, 2], [0, 1, 2], [0, 1, 2], [0, 1, 2]])
final_perm = offset + extended_perm[:, None]
final_perm = final_perm.view(-1)
else:
final_perm = extended_perm
permuted_img = partitioned_img[:, final_perm, :, :]
permuted_img = rebuild_tensor(permuted_img, base, tile_size=inputs.size(2) // 2, offset=inputs.size(2) // 2)
img = Variable(permuted_img).cuda()
outputs = model(img)
outputs = outputs.view(outputs.shape[0], -1)
target = target.view(target.shape[0], -1)
scores = torch.mean((target - outputs) ** 2, dim=1)
scores /= torch.tensor(perm_cost[idx])
min_score = torch.min(scores, min_score)
max_score = torch.max(scores, max_score)
avg_score += scores
idx += 1
avg_score = avg_score / num_perm
label_score_avg += list(zip(labels.cpu().data.numpy().tolist(), avg_score.cpu().data.numpy().tolist()))
label_score_max += list(zip(labels.cpu().data.numpy().tolist(), max_score.cpu().data.numpy().tolist()))
label_score_min += list(zip(labels.cpu().data.numpy().tolist(), min_score.cpu().data.numpy().tolist()))
label_scores = {'max': label_score_max, 'min': label_score_min, 'avg': label_score_avg}
AUC = dict()
for key, label_score in label_scores.items():
labels, scores = zip(*label_score)
labels = np.array(labels)
indx1 = labels == target_class
indx2 = labels != target_class
labels[indx1] = 1
labels[indx2] = 0
scores = np.array(scores)
fpr, tpr, thresholds = roc_curve(labels, scores, pos_label=0)
roc_auc = auc(fpr, tpr)
roc_auc = round(roc_auc, 4)
AUC[key] = roc_auc
return AUC
def get_avg_val_error_per_permutation(model, permutation_list, val_dataloader):
permutation_cost = []
for ind, perm in enumerate(permutation_list):
avg_score = 0
for data in val_dataloader:
inputs = data[0]
orig_img = inputs
target = orig_img
target = Variable(target).cuda()
partitioned_img, base = split_tensor(inputs, tile_size=inputs.size(2) // 2, offset=inputs.size(2) // 2)
extended_perm = torch.tensor(perm) * inputs.size(1)
if inputs.size(1) == 3:
offset = torch.tensor([[0, 1, 2], [0, 1, 2], [0, 1, 2], [0, 1, 2]])
final_perm = offset + extended_perm[:, None]
final_perm = final_perm.view(-1)
else:
final_perm = extended_perm
permuted_img = partitioned_img[:, final_perm, :, :]
permuted_img = rebuild_tensor(permuted_img, base, tile_size=inputs.size(2) // 2, offset=inputs.size(2) // 2)
img = Variable(permuted_img).cuda()
outputs = model(img)
outputs = outputs.view(outputs.shape[0], -1)
target = target.view(target.shape[0], -1)
scores = torch.mean((target - outputs) ** 2)
avg_score += scores.item()
permutation_cost.append(avg_score)
return permutation_cost
def main():
args = parser.parse_args()
config = get_config(args.config)
n_channel = config['n_channel']
normal_class = config["normal_class"]
dataset_name = config['dataset_name']
checkpoint_path = "outputs/{}/{}/checkpoints/".format(dataset_name, normal_class)
if dataset_name != "MVTec":
target_class = normal_class
else:
mvtec_good_dict = {'bottle': 3, 'cable': 5, 'capsule': 2, 'carpet': 2,
'grid': 3, 'hazelnut': 2, 'leather': 4, 'metal_nut': 3, 'pill': 5,
'screw': 0, 'tile': 2, 'toothbrush': 1, 'transistor': 3, 'wood': 2,
'zipper': 4
}
target_class = mvtec_good_dict[normal_class]
_, val_dataloader, test_dataloader = load_data(config)
model = UNet(n_channel, n_channel).cuda()
model.load_state_dict(torch.load(checkpoint_path + '{}.pth'.format(str(config['last_epoch']))))
permutation_list = get_all_permutations()
perm_cost = get_avg_val_error_per_permutation(model, permutation_list, val_dataloader)
auc_dict = test(model, target_class, permutation_list, perm_cost, test_dataloader)
print(auc_dict)
if __name__ == '__main__':
main()