-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdatasets.py
192 lines (149 loc) · 6.35 KB
/
datasets.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
import cv2
import h5py
import numpy as np
import scipy.io as scio
import torch
from PIL import Image
from torchvision import transforms
import settings
np.random.seed(1)
if settings.DATASET == "MIRFlickr":
label_set = scio.loadmat(settings.LABEL_DIR)
label_set = np.array(label_set['LAll'], dtype=np.float)
txt_set = scio.loadmat(settings.TXT_DIR)
txt_set = np.array(txt_set['YAll'], dtype=np.float)
first = True
for label in range(label_set.shape[1]):
index = np.where(label_set[:, label] == 1)[0]
N = index.shape[0]
perm = np.random.permutation(N)
index = index[perm]
if first:
test_index = index[:160]
train_index = index[160:160 + 800]
first = False
else:
ind = np.array([i for i in list(index) if i not in (list(train_index) + list(test_index))])
test_index = np.concatenate((test_index, ind[:80]))
train_index = np.concatenate((train_index, ind[80:80 + 400]))
database_index = np.array([i for i in list(range(label_set.shape[0])) if i not in list(test_index)])
if train_index.shape[0] < 10000:
pick = np.array([i for i in list(database_index) if i not in list(train_index)])
N = pick.shape[0]
perm = np.random.permutation(N)
pick = pick[perm]
res = 10000 - train_index.shape[0]
train_index = np.concatenate((train_index, pick[:res]))
indexTest = test_index
indexDatabase = database_index
indexTrain = train_index
mir_train_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.Resize(256),
transforms.RandomCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
mir_test_transform = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
txt_feat_len = txt_set.shape[1]
class MIRFlickr(torch.utils.data.Dataset):
def __init__(self, transform=None, target_transform=None, train=True, database=False):
self.transform = transform
self.target_transform = target_transform
if train:
self.train_labels = label_set[indexTrain]
self.train_index = indexTrain
self.txt = txt_set[indexTrain]
elif database:
self.train_labels = label_set[indexDatabase]
self.train_index = indexDatabase
self.txt = txt_set[indexDatabase]
else:
self.train_labels = label_set[indexTest]
self.train_index = indexTest
self.txt = txt_set[indexTest]
def __getitem__(self, index):
mirflickr = h5py.File(settings.IMG_DIR, 'r', libver='latest', swmr=True)
img, target = mirflickr['IAll'][self.train_index[index]], self.train_labels[index]
img = Image.fromarray(np.transpose(img, (2, 1, 0)))
mirflickr.close()
txt = self.txt[index]
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, txt, target, index
def __len__(self):
return len(self.train_labels)
if settings.DATASET == "NUSWIDE":
label_set = scio.loadmat(settings.LABEL_DIR)
label_set = np.array(label_set['LAll'], dtype=np.float)
txt_file = h5py.File(settings.TXT_DIR, 'r')
txt_set = np.array(txt_file['YAll']).transpose()
txt_file.close()
first = True
for label in range(label_set.shape[1]):
index = np.where(label_set[:, label] == 1)[0]
N = index.shape[0]
perm = np.random.permutation(N)
index = index[perm]
if first:
test_index = index[:200]
train_index = index[200:1200]
first = False
else:
ind = np.array([i for i in list(index) if i not in (list(train_index) + list(test_index))])
test_index = np.concatenate((test_index, ind[:200]))
train_index = np.concatenate((train_index, ind[200:1200]))
database_index = np.array([i for i in list(range(label_set.shape[0])) if i not in list(test_index)])
indexTest = test_index
indexDatabase = database_index
indexTrain = train_index
nus_train_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.Resize(256),
transforms.RandomCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
nus_test_transform = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
txt_feat_len = txt_set.shape[1]
class NUSWIDE(torch.utils.data.Dataset):
def __init__(self, transform=None, target_transform=None, train=True, database=False):
self.transform = transform
self.target_transform = target_transform
if train:
self.train_labels = label_set[indexTrain]
self.train_index = indexTrain
self.txt = txt_set[indexTrain]
elif database:
self.train_labels = label_set[indexDatabase]
self.train_index = indexDatabase
self.txt = txt_set[indexDatabase]
else:
self.train_labels = label_set[indexTest]
self.train_index = indexTest
self.txt = txt_set[indexTest]
def __getitem__(self, index):
nuswide = h5py.File(settings.IMG_DIR, 'r', libver='latest', swmr=True)
img, target = nuswide['IAll'][self.train_index[index]], self.train_labels[index]
img = Image.fromarray(np.transpose(img, (2, 1, 0)))
nuswide.close()
txt = self.txt[index]
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, txt, target, index
def __len__(self):
return len(self.train_labels)