-
Notifications
You must be signed in to change notification settings - Fork 10
/
Copy pathget_model_features.py
158 lines (108 loc) · 5.3 KB
/
get_model_features.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
import torch
import random
import torchvision
import torchvision.transforms as transforms
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import DHR_Net as models
import numpy as np
import pickle
import os
from PIL import Image
def epoch_anomalous(net,save_path,root):
net.eval()
with torch.no_grad():
for folder in os.listdir(os.path.join(root,"open_set")):
count=0
for file_name in os.listdir(os.path.join(root,"open_set",str(folder))):
if(count>=120):
break
count = count + 1
image = Image.open(os.path.join(root,"open_set",str(folder),file_name)).convert("RGB")
image = transform_test(image)
image = torch.unsqueeze(image,0)
image=image.cuda(non_blocking=True)
logits, _, latent = net(image)
squeezed_latent = []
squeezed_latent.append(logits)
for layer in latent:
m = nn.AdaptiveAvgPool2d((1,1))
new_layer = torch.squeeze(m(layer))
squeezed_latent.append(new_layer)
feature = torch.cat(squeezed_latent,1)
save_name = file_name.split(".")[0]
print(save_name)
np.save(os.path.join(save_path,"open_set",str(folder),save_name+".npy"),feature.cpu().data.numpy(),allow_pickle=False)
def epoch_train(net,save_path,root):
net.eval()
with torch.no_grad():
for folder in os.listdir(os.path.join(root,"train")):
for file_name in os.listdir(os.path.join(root,"train",str(folder))):
image = Image.open(os.path.join(root,"train",str(folder),file_name)).convert("RGB")
image = transform_test(image)
image = torch.unsqueeze(image,0)
image=image.cuda(non_blocking=True)
logits, _, latent = net(image)
squeezed_latent = []
squeezed_latent.append(logits)
for layer in latent:
m = nn.AdaptiveAvgPool2d((1,1))
new_layer = torch.squeeze(m(layer))
squeezed_latent.append(new_layer)
feature = torch.cat(squeezed_latent,1)
save_name = file_name.split(".")[0]
print(save_name)
np.save(os.path.join(save_path,"train",str(folder),save_name+".npy"),feature.cpu().data.numpy(),allow_pickle=False)
def epoch_val(net,save_path,root):
net.eval()
with torch.no_grad():
for folder in os.listdir(os.path.join(root,"val")):
for file_name in os.listdir(os.path.join(root,"val",str(folder))):
image = Image.open(os.path.join(root,"val",str(folder),file_name)).convert("RGB")
image = transform_test(image)
image = torch.unsqueeze(image,0)
image=image.cuda(non_blocking=True)
logits, _, latent = net(image)
squeezed_latent = []
squeezed_latent.append(logits)
for layer in latent:
m = nn.AdaptiveAvgPool2d((1,1))
new_layer = torch.squeeze(m(layer))
squeezed_latent.append(new_layer)
feature = torch.cat(squeezed_latent,1)
save_name = file_name.split(".")[0]
print(save_name)
np.save(os.path.join(save_path,"val",str(folder),save_name+".npy"),feature.cpu().data.numpy(),allow_pickle=False)
def get_args():
parser = argparse.ArgumentParser(description='Get activation vectors')
parser.add_argument('--dataset_dir',default="./data/cifar10",type=str,help="Number of members in ensemble")
parser.add_argument('--num_classes',default=6,type=int,help="Number of classes in dataset")
parser.add_argument('--means',nargs='+',default=[0.4914, 0.4822, 0.4465],type=float,help="channelwise means for normalization")
parser.add_argument('--stds',nargs='+',default=[0.2023, 0.1994, 0.2010],type=float,help="channelwise std for normalization")
parser.add_argument('--save_path',default="./saved_features/cifar10",type=str,help="Path to save the ensemble weights")
parser.add_argument('--load_path',default="./saved_models/cifar10/50.pth",type=str,help="Path to save the ensemble weights")
parser.set_defaults(argument=True)
return parser.parse_args()
def main():
args = get_args()
seed = 222
np.random.seed(seed)
torch.manual_seed(seed)
random.seed(seed)
num_classes = args.num_classes
print("Num classes "+str(num_classes))
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(args.means, args.stds),
])
root = args.dataset_dir
net = models.DHRNet(num_classes)
checkpoint = torch.load(args.load_path,map_location="cpu")
net.load_state_dict(checkpoint)
net.cuda()
epoch_train(net,args.save_path,root)
epoch_val(net,args.save_path,root)
epoch_openset(net,args.save_path,root)
if __name__=="__main__":
main()