-
Notifications
You must be signed in to change notification settings - Fork 5
/
dataset.py
148 lines (129 loc) · 6.08 KB
/
dataset.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
import torch.utils.data as data
import numpy as np
from utils import process_feat, get_rgb_list_file
import torch
from torch.utils.data import DataLoader
torch.set_default_tensor_type('torch.cuda.FloatTensor')
class Dataset(data.Dataset):
def __init__(self, args, is_normal=True, transform=None, test_mode=False):
self.modality = args.modality
self.emb_folder = args.emb_folder
self.is_normal = is_normal
self.dataset = args.dataset
self.feature_size = args.feature_size
if args.test_rgb_list is None:
_, self.rgb_list_file = get_rgb_list_file(args.dataset, test_mode)
else:
self.rgb_list_file = args.test_rgb_list
# deal with different I3D feature version
if 'v2' in self.dataset:
self.feat_ver = 'v2'
elif 'v3' in self.dataset:
self.feat_ver = 'v3'
else:
self.feat_ver = 'v1'
self.tranform = transform
self.test_mode = test_mode
self._parse_list()
self.num_frame = 0
self.labels = None
def _parse_list(self):
self.list = list(open(self.rgb_list_file))
if self.test_mode is False: # list for training would need to be ordered from normal to abnormal
if 'shanghai' in self.dataset:
if self.is_normal:
self.list = self.list[63:]
print('normal list for shanghai tech')
else:
self.list = self.list[:63]
print('abnormal list for shanghai tech')
elif 'ucf' in self.dataset:
if self.is_normal:
self.list = self.list[810:]
print('normal list for ucf')
else:
self.list = self.list[:810]
print('abnormal list for ucf')
elif 'violence' in self.dataset:
if self.is_normal:
self.list = self.list[1904:]
print('normal list for violence')
else:
self.list = self.list[:1904]
print('abnormal list for violence')
elif 'ped2' in self.dataset:
if self.is_normal:
self.list = self.list[6:]
print('normal list for ped2', len(self.list))
else:
self.list = self.list[:6]
print('abnormal list for ped2', len(self.list))
elif 'TE2' in self.dataset: # 注意index从0开始，而pycharm行号从1开始
if self.is_normal:
self.list = self.list[23:]
print('normal list for TE2', len(self.list))
else:
self.list = self.list[:23]
print('abnormal list for TE2', len(self.list))
else:
raise Exception("Dataset undefined!!!")
def __getitem__(self, index):
label = self.get_label() # get video level label 0/1
i3d_path = self.list[index].strip('\n')
if self.feat_ver == 'v2':
i3d_path = i3d_path.replace('i3d_v1', 'i3d_v2')
elif self.feat_ver == 'v3':
i3d_path = i3d_path.replace('i3d_v1', 'i3d_v3')
features = np.load(i3d_path, allow_pickle=True)
features = np.array(features, dtype=np.float32)
if 'ucf' in self.dataset:
text_path = "save/Crime/" + self.emb_folder + "/" + i3d_path.split("/")[-1][:-7]+"emb.npy"
elif 'shanghai' in self.dataset:
text_path = "save/Shanghai/" + self.emb_folder + "/" + i3d_path.split("/")[-1][:-7]+"emb.npy"
elif 'violence' in self.dataset:
text_path = "save/Violence/" + self.emb_folder + "/" + i3d_path.split("/")[-1][:-7]+"emb.npy"
elif 'ped2' in self.dataset:
text_path = "save/UCSDped2/" + self.emb_folder + "/" + i3d_path.split("/")[-1][:-7]+"emb.npy"
elif 'TE2' in self.dataset:
text_path = "save/TE2/" + self.emb_folder + "/" + i3d_path.split("/")[-1][:-7]+"emb.npy"
else:
raise Exception("Dataset undefined!!!")
text_features = np.load(text_path, allow_pickle=True)
text_features = np.array(text_features, dtype=np.float32) # [snippet no., 768]
# assert features.shape[0] == text_features.shape[0]
if self.feature_size == 1024:
text_features = np.tile(text_features, (5, 1, 1)) # [10,snippet no.,768]
elif self.feature_size == 2048:
text_features = np.tile(text_features, (10, 1, 1)) # [10,snippet no.,768]
else:
raise Exception("Feature size undefined!!!")
if self.tranform is not None:
features = self.tranform(features)
if self.test_mode:
text_features = text_features.transpose(1, 0, 2) # [snippet no.,10,768]
return features, text_features
else:
# process 10-cropped snippet feature
features = features.transpose(1, 0, 2) # [snippet no., 10, 2048] -> [10, snippet no., 2048]
divided_features = []
for feature in features: # loop 10 times
feature = process_feat(feature, 32) # divide a video into 32 segments/snippets/clips
divided_features.append(feature)
divided_features = np.array(divided_features, dtype=np.float32) # [10,32,2048]
div_feat_text = []
for text_feat in text_features:
text_feat = process_feat(text_feat, 32) # [32,768]
div_feat_text.append(text_feat)
div_feat_text = np.array(div_feat_text, dtype=np.float32)
assert divided_features.shape[1] == div_feat_text.shape[1], str(self.test_mode) + "\t" + str(divided_features.shape[1]) + "\t" + div_feat_text.shape[1]
return divided_features, div_feat_text, label
def get_label(self):
if self.is_normal:
label = torch.tensor(0.0)
else:
label = torch.tensor(1.0)
return label
def __len__(self):
return len(self.list)
def get_num_frames(self):
return self.num_frame