-
Notifications
You must be signed in to change notification settings - Fork 1
/
data.py
263 lines (214 loc) · 8.14 KB
/
data.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
255
256
257
258
259
260
261
262
263
import os
import pandas as pd
import torchvision.transforms as transforms
from operator import itemgetter
from torch.utils.data import Dataset
from PIL import Image
import torch
import random
MEAN = [0.485, 0.456, 0.406]
STD = [0.229, 0.224, 0.225]
def read_csv(csv_path):
return pd.read_csv(csv_path, header=None, index_col=False)
class DATA(Dataset):
def __init__(self, args, mode='train'):
''' set up basic parameters for dataset '''
self.mode = mode
data_dir = args.data_dir
csv_path = os.path.join(data_dir, mode + '.csv')
img_dir = os.path.join(data_dir, 'imgs')
data_F = read_csv(csv_path)
self.imgs_names = read_csv(csv_path)[1]
for i in range(len(data_F)):
data_F[1][i] = os.path.join(img_dir, data_F[1][i])
# Group Labels for the training set
if mode == 'train':
data_S = data_F.values.tolist()
data_S.sort(key=itemgetter(0))
last_num = -1
idx = -1
# Change labels to go from 0 to number of different labels
for i in range(len(data_S)):
if data_S[i][0] != last_num:
idx += 1
last_num = data_S[i][0]
data_S[i][0] = idx
last_num = -1
cnt = 0
num = args.label_group - 1
self.data = []
for i in range(len(data_S)):
if(data_S[i][0] == last_num) & (cnt < num):
cnt += 1
else:
self.data.append([])
cnt = 0
if i == (len(data_S)-1):
if cnt != num:
for e in range(num-cnt):
self.data[-1].append(data_S[i-e-1])
elif (data_S[i][0] != data_S[i+1][0]) & (cnt != num):
for e in range(num-cnt):
self.data[-1].append(data_S[i-e-1])
self.data[-1].append(data_S[i])
last_num = data_S[i][0]
else:
self.data = data_F
img_size = 224
''' set up image trainsform '''
self.transform = transforms.Compose([
transforms.Resize(img_size),
#transforms.RandomHorizontalFlip(p=0.5),
transforms.CenterCrop(img_size),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD),
])
self.transform_t = transforms.Compose([
transforms.Resize(img_size),
#transforms.CenterCrop(img_size),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD)
])
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
''' get data '''
if self.mode == 'train':
images = []
for i in range(len(self.data[idx])):
cls, img_path = self.data[idx][i][0], self.data[idx][i][1]
img = Image.open(img_path).convert('RGB')
img_t = self.transform(img)
images.append(img_t)
images = torch.stack(images)
else:
img = Image.open(self.data[1][idx]).convert('RGB')
images = self.transform_t(img)
cls = self.imgs_names[idx]
''' read image '''
return images, cls
# Dataloader with random geting of images from the same label, slight difference.
class DATA2(Dataset):
def __init__(self, args, mode='train'):
''' set up basic parameters for dataset '''
self.args = args
self.mode = mode
data_dir = args.data_dir
csv_path = os.path.join(data_dir, mode + '.csv')
img_dir = os.path.join(data_dir, 'imgs')
data_F = read_csv(csv_path)
self.imgs_names = read_csv(csv_path)[1]
for i in range(len(data_F)):
data_F[1][i] = os.path.join(img_dir, data_F[1][i])
# Group Labels for the training set
if mode == 'train':
data_S = data_F.values.tolist()
data_S.sort(key=itemgetter(0))
# Change labels to go from 0 to number of different labels
last_num = -1
idx = 0
for i in range(len(data_S)):
if data_S[i][0] != last_num:
last_num = data_S[i][0]
idx += 5
data_S[i][0] = idx
last_num = -1
self.data = []
for i in range(len(data_S)):
if (data_S[i][0] != last_num):
self.data.append([])
self.data[-1].append(data_S[i])
last_num = data_S[i][0]
else:
self.data = data_F
img_size = 224
''' set up image trainsform '''
self.transform = transforms.Compose([
transforms.Resize(img_size),
#transforms.RandomHorizontalFlip(p=0.5),
#transforms.Pad(5),
transforms.RandomCrop((img_size, img_size)),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD),
])
self.transform_t = transforms.Compose([
transforms.Resize(img_size),
transforms.CenterCrop(img_size),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD)
])
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
''' get data '''
if self.mode == 'train':
images = []
random_ids = random.sample(range(0, len(self.data[idx])), self.args.label_group)
#rand = bool(random.getrandbits(1))
for i in random_ids:
cls, img_path = self.data[idx][i][0], self.data[idx][i][1]
img = Image.open(img_path).convert('RGB')
img_t = self.transform(img)
#img_t = trans(img, rand)
images.append(img_t)
images = torch.stack(images)
else:
img = Image.open(self.data[1][idx]).convert('RGB')
images = self.transform_t(img)
cls = self.imgs_names[idx]
''' read image '''
return images, cls
#Dataloader with random geting of images from the same label, slight difference.
class DATA_un(Dataset):
def __init__(self, args):
''' set up basic parameters for dataset '''
self.args = args
data_dir = args.data_dir
csv_path = os.path.join(data_dir, 'train_unlabeled.csv')
self.img_dir = os.path.join(data_dir, 'imgs')
self.imgs_names = read_csv(csv_path)[0]
img_size = 224
''' set up image transform '''
self.transform = transforms.Compose([
transforms.Resize(img_size),
transforms.CenterCrop(img_size),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD)
])
def __len__(self):
return len(self.imgs_names)
def __getitem__(self, idx):
img = Image.open(os.path.join(self.img_dir, self.imgs_names[idx])).convert('RGB')
images = self.transform(img)
''' read image '''
return images, self.imgs_names[idx]
# Dataloader with random geting of images from the same label, slight difference.
class DATA_final(Dataset):
def __init__(self, args, mode = 'query'):
''' set up basic parameters for dataset '''
self.args = args
if mode == 'query':
csv = args.query_dir
else:
csv = args.gallery_dir
self.imgs_dirs = read_csv(csv)[0]
self.imgs_names = read_csv(csv)[0]
for i in range(len(self.imgs_names)):
self.imgs_dirs[i] = os.path.join(args.img_dir, self.imgs_dirs[i])
img_size = 224
''' set up image trainsform '''
self.transform = transforms.Compose([
transforms.Resize(img_size),
transforms.CenterCrop(img_size),
transforms.ToTensor(),
transforms.Normalize(MEAN, STD)
])
def __len__(self):
return len(self.imgs_dirs)
def __getitem__(self, idx):
''' get data '''
img = Image.open(self.imgs_dirs[idx]).convert('RGB')
images = self.transform(img)
cls = self.imgs_names[idx]
''' read image '''
return images, cls