-
Notifications
You must be signed in to change notification settings - Fork 25
/
dataset.py
361 lines (298 loc) · 13 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
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
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
from torch.utils.data import Dataset
from torchvision import transforms
from PIL import Image
import tarfile
import hashlib
import os
import os.path
import sys
import shutil
import tempfile
import errno
import numpy as np
import csv
import math
try:
from requests.utils import urlparse
import requests.get as urlopen
requests_available = True
except ImportError:
from urllib.request import urlopen
from urllib.parse import urlparse
requests_available = False
try:
from tqdm import tqdm
except ImportError:
tqdm = None
def default_loader(path):
return Image.open(path).convert('RGB')
class FashionAI(Dataset):
urls = [
'https://en.myluo.cn/packages/fashionAI_attributes_test_a_20180222.tar',
'https://en.myluo.cn/packages/fashionAI_attributes_train_20180222.tar',
'https://en.myluo.cn/packages/warm_up_train_20180201.tar',
]
hashs = [
'765095dcb83edb0af137e8295244c044137b47feddc3b552fd03d1165c57f495',
'20ddae0deb5544b6ccdda7df46fd2e0c7756bba2e7ed4c9a23c2c133ede138c6',
'6c6957250e11cc146634eef87cd044fa9979b2d28355c24a059c9ce05b85a7cb',
]
base_folder = 'datasets'
train_folder = 'base'
warm_folder = 'web'
rank_folder = 'rank'
data_folder = 'Images'
label_folder = 'Annotations'
rank_label_folder = 'Tests'
train_label = 'label.csv'
rank_label = 'question.csv'
shuffle_file = '_shuffle.npy'
train_data_file = '_train_data.npy'
train_label_file = '_train_label.npy'
test_data_file = '_test_data.npy'
test_label_file = '_test_label.npy'
rank_data_file = '_rank_data.npy'
rank_index_file = '_rank_index.npy'
ms_file = '_ms.npy'
AttrKey = {
'coat_length_labels':8,
'collar_design_labels':5,
'lapel_design_labels':5,
'neck_design_labels':5,
'neckline_design_labels':10,
'pant_length_labels':6,
'skirt_length_labels':6,
'sleeve_length_labels':9,
}
def __init__(self, root, attribute, split=0.8, data_type='train', reset=False, ci=False,
transform=None, target_transform=None, loader=default_loader):
self.root = os.path.expanduser(root)
self.attribute = attribute
self.transform = transform
self.target_transform = target_transform
self.data_type = data_type
self.loader = loader
self.split = split
self.reset = reset
self.ci = ci
self.depth = 3
self.width = 224
self.height = 224
if self.attribute not in self.AttrKey.keys():
raise RuntimeError('Attribute not found.')
else:
self.nb_classes = self.AttrKey[self.attribute]
if self.split <= 0 or self.split >= 1:
self.split = 0.8
self.download()
label_file = os.path.join(self.root, self.base_folder, self.train_folder, self.label_folder,
self.train_label)
csvdata = []
with open(label_file) as f:
reader = csv.reader(f)
for row in reader:
if row[1] == self.attribute:
csvdata.append(row)
shuffle_file = os.path.join(self.root, self.base_folder, self.train_folder, self.label_folder,
self.attribute + self.shuffle_file)
if not os.path.exists(shuffle_file) or (self.data_type == 'train' and self.reset):
shuffle = list(range(len(csvdata)))
np.random.shuffle(shuffle)
np.save(shuffle_file, shuffle)
self.reset = True
else:
shuffle = np.load(shuffle_file)
ms_file = os.path.join(self.root, self.base_folder, self.train_folder, self.data_folder,
self.attribute + self.ms_file)
if os.path.exists(ms_file):
ms = np.load(ms_file)
self.mean = ms[0]
self.std = ms[1]
if self.transform is None:
self.transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(self.mean, self.std)
])
else:
self.reset = True
if self.data_type == 'train':
data_file = os.path.join(self.root, self.base_folder, self.train_folder, self.data_folder,
self.attribute + self.train_data_file)
label_file = os.path.join(self.root, self.base_folder, self.train_folder, self.data_folder,
self.attribute + self.train_label_file)
if not self.reset and os.path.exists(data_file) and os.path.exists(label_file):
self.train_data = np.load(data_file)
self.train_labels = np.load(label_file)
return
self.train_data = []
self.train_labels = []
count = 0
for row in shuffle[:math.floor(self.split * len(shuffle))]:
image_file = os.path.join(self.root, self.base_folder, self.train_folder, csvdata[row][0])
self.train_data.append(np.uint8(np.array(self.loader(image_file).resize((self.width, self.height))).tolist()))
self.train_labels.append(csvdata[row][2].find('y'))
count += 1
if self.ci and count == 80:
break
self.train_data = np.array(self.train_data)
#self.train_labels = np.eye(self.nb_classes, dtype=int)[np.array(self.train_labels).reshape(-1)]
rdata = self.train_data.reshape((count * self.width * self.height, self.depth))
self.mean = tuple(np.mean(rdata, 0))
self.std = tuple(np.std(rdata, 0))
if self.transform is None:
self.transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(self.mean, self.std)
])
np.save(data_file, self.train_data)
np.save(label_file, self.train_labels)
np.save(ms_file, [self.mean, self.std])
#print(len(self.train_data.shape))
#print(self.train_labels)
elif self.data_type == 'test':
data_file = os.path.join(self.root, self.base_folder, self.train_folder, self.data_folder,
self.attribute + self.test_data_file)
label_file = os.path.join(self.root, self.base_folder, self.train_folder, self.data_folder,
self.attribute + self.test_label_file)
if not self.reset and os.path.exists(data_file) and os.path.exists(label_file):
self.test_data = np.load(data_file)
self.test_labels = np.load(label_file)
return
self.test_data = []
self.test_labels = []
count = 0
for row in shuffle[math.ceil(self.split * len(shuffle)):]:
image_file = os.path.join(self.root, self.base_folder, self.train_folder, csvdata[row][0])
self.test_data.append(np.uint8(np.array(self.loader(image_file).resize((self.width, self.height))).tolist()))
self.test_labels.append(csvdata[row][2].find('y'))
count += 1
if self.ci and count == 20:
break
self.test_data = np.array(self.test_data)
#self.test_labels = np.eye(self.nb_classes, dtype=int)[np.array(self.test_labels).reshape(-1)]
np.save(data_file, self.test_data)
np.save(label_file, self.test_labels)
#print(self.test_data.shape)
#print(self.test_labels)
elif self.data_type == 'eval':
data_file = os.path.join(self.root, self.base_folder, self.rank_folder, self.data_folder,
self.attribute + self.rank_data_file)
index_file = os.path.join(self.root, self.base_folder, self.rank_folder, self.data_folder,
self.attribute + self.rank_index_file)
if not self.reset and os.path.exists(data_file) and os.path.exists(index_file):
self.eval_data = np.load(data_file)
self.eval_index = np.load(index_file)
return
self.eval_data = []
self.eval_index = []
label_file = os.path.join(self.root, self.base_folder, self.rank_folder, self.rank_label_folder,
self.rank_label)
count = 0
with open(label_file) as f:
reader = csv.reader(f)
for row in reader:
if row[1] == self.attribute:
image_file = os.path.join(self.root, self.base_folder, self.rank_folder, row[0])
self.eval_data.append(np.uint8(np.array(self.loader(image_file).resize((self.width, self.height))).tolist()))
self.eval_index.append(row[0])
count += 1
if self.ci and count == 20:
break
self.eval_data = np.array(self.eval_data)
np.save(data_file, self.eval_data)
np.save(index_file, self.eval_index)
#print(self.test_data.shape)
def __getitem__(self, index):
if self.data_type == 'train':
img, target = self.train_data[index], self.train_labels[index]
elif self.data_type == 'test':
img, target = self.test_data[index], self.test_labels[index]
elif self.data_type == 'eval':
img, target = self.eval_data[index], self.eval_index[index]
img = Image.fromarray(img)
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
if self.data_type == 'train':
return len(self.train_data)
elif self.data_type == 'test':
return len(self.test_data)
elif self.data_type == 'eval':
return len(self.eval_data)
def _check_exists(self):
return os.path.exists(os.path.join(self.root, self.base_folder, self.train_folder)) and \
os.path.exists(os.path.join(self.root, self.base_folder, self.warm_folder)) and \
os.path.exists(os.path.join(self.root, self.base_folder, self.rank_folder))
def _download_url_to_file(self, url, dst, hash_prefix, progress=True):
u = urlopen(url)
if requests_available:
file_size = int(u.headers["Content-Length"])
u = u.raw
else:
meta = u.info()
if hasattr(meta, 'getheaders'):
file_size = int(meta.getheaders("Content-Length")[0])
else:
file_size = int(meta.get_all("Content-Length")[0])
f = tempfile.NamedTemporaryFile(delete=False)
try:
sha256 = hashlib.sha256()
with tqdm(total=file_size, disable=not progress) as pbar:
while True:
buffer = u.read(8192)
if len(buffer) == 0:
break
f.write(buffer)
sha256.update(buffer)
pbar.update(len(buffer))
f.close()
digest = sha256.hexdigest()
if digest[:len(hash_prefix)] != hash_prefix:
raise RuntimeError('invalid hash value (expected "{}", got "{}")'
.format(hash_prefix, digest))
shutil.move(f.name, dst)
finally:
f.close()
if os.path.exists(f.name):
os.remove(f.name)
def download(self):
if self._check_exists():
return
try:
os.makedirs(os.path.join(self.root, self.base_folder))
except OSError as e:
if e.errno == errno.EEXIST:
pass
else:
raise
for url, hash in zip(self.urls, self.hashs):
filename = url.rpartition('/')[2]
file_folder = os.path.join(self.root, self.base_folder)
file_path = os.path.join(file_folder, filename)
sys.stderr.write('Downloading: "{}" to {}\n'.format(url, file_path))
self._download_url_to_file(url, file_path, hash)
with tarfile.open(file_path) as tar_f:
tar_f.extractall(path=file_folder)
if tqdm is None:
class tqdm(object):
def __init__(self, total, disable=False):
self.total = total
self.disable = disable
self.n = 0
def update(self, n):
if self.disable:
return
self.n += n
sys.stderr.write("\r{0:.1f}%".format(100 * self.n / float(self.total)))
sys.stderr.flush()
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
if self.disable:
return
sys.stderr.write('\n')
if __name__ == "__main__":
FashionAI('./', 'coat_length_labels')