-
Notifications
You must be signed in to change notification settings - Fork 1
/
data.py
387 lines (311 loc) · 11.4 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
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
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
print("start data import")
from tensorflow.python.platform import flags
from imageio import imread
import tensorflow as tf
import io
import lmdb
import logging
import tqdm
import numpy
from PIL import Image
import json
from torch.utils.data import Dataset
import pickle
import os.path as osp
import os
import numpy as np
import time
from PIL import Image
#from scipy.misc import imread, imresize
#from skimage import io
from skimage.color import rgb2gray
from torchvision.datasets import CIFAR10, MNIST, SVHN, CIFAR100, ImageFolder, LSUNClass, LSUN
from torchvision import transforms
import torch
import torchvision
import pandas as pd
from imageio import imwrite
from absl import flags
import errno
import codecs
from torch.utils import data
import random
print("end data import")
def cutout(mask_color=(0, 0, 0)):
mask_size_half = FLAGS.cutout_mask_size // 2
offset = 1 if FLAGS.cutout_mask_size % 2 == 0 else 0
def _cutout(image):
image = np.asarray(image).copy()
if np.random.random() > FLAGS.cutout_prob:
return image
h, w = image.shape[:2]
if FLAGS.cutout_inside:
cxmin, cxmax = mask_size_half, w + offset - mask_size_half
cymin, cymax = mask_size_half, h + offset - mask_size_half
else:
cxmin, cxmax = 0, w + offset
cymin, cymax = 0, h + offset
cx = np.random.randint(cxmin, cxmax)
cy = np.random.randint(cymin, cymax)
xmin = cx - mask_size_half
ymin = cy - mask_size_half
xmax = xmin + FLAGS.cutout_mask_size
ymax = ymin + FLAGS.cutout_mask_size
xmin = max(0, xmin)
ymin = max(0, ymin)
xmax = min(w, xmax)
ymax = min(h, ymax)
image[:, ymin:ymax, xmin:xmax] = np.array(mask_color)[:, None, None]
return image
return _cutout
class CelebAHQ(Dataset):
def __init__(self, cond_idx=1, filter_idx=0):
self.path = "/datasets01/celebAHQ/081318/imgHQ{:05}.npy"
self.labels = pd.read_csv("/private/home/yilundu/list_attr_celeba.txt", sep="\s+", skiprows=1)
self.hq_labels = pd.read_csv("/private/home/yilundu/image_list.txt", sep="\s+")
self.cond_idx = cond_idx
self.filter_idx = filter_idx
def __len__(self):
return self.hq_labels.shape[0]
def __getitem__(self, index):
info = self.hq_labels.iloc[index]
info = self.labels.iloc[info.orig_idx]
path = self.path.format(index)
im = np.load(path)
im = im[0].transpose((1, 2, 0))
image_size = 128
im = imresize(im, (image_size, image_size))
im = im / 256
im = im + np.random.uniform(0, 1 / 256., im.shape)
label = int(info.iloc[self.cond_idx])
if label == -1:
label = 0
label = np.eye(2)[label]
im_corrupt = np.random.uniform(
0, 1, size=(image_size, image_size, 3))
return im_corrupt, im, label
class ImageNet(Dataset):
def __init__(self, cond_idx=1, filter_idx=0):
self.path = "/raid/common/imagenet-raw/train"
#print(os.listdir(self.path))
self.folders = [osp.join(self.path, d) for d in os.listdir(self.path)]
self.images = []
self.labels = []
for i, folder in enumerate(self.folders):
im_path = [osp.join(folder, im) for im in os.listdir(folder)]
self.images.extend(im_path)
self.labels.extend([i] * len(im_path))
def __len__(self):
return len(self.images)
def __getitem__(self, index):
path = self.images[index]
im = imread(path)
if len(im.shape) == 2:
im = np.tile(im[:, :, None], (1, 1, 3))
else:
im = im[:, :, :3]
image_size = 64
im = numpy.array(Image.fromarray(im).resize((image_size,image_size)))
#im = imresize(im, (image_size, image_size))
im = im / 256
im = im + np.random.uniform(0, 1 / 256., im.shape)
im = im.transpose((2, 0, 1))
im = (im-0.5)/0.5
#im_corrupt = np.random.uniform(0, 1, size=(image_size, image_size, 3))
label = np.eye(1000)[self.labels[index]]
return im,label
class LSUNBed(Dataset):
def __init__(self, cond_idx=1, filter_idx=0):
self.path = "/home/aiops/allanguo"
lmdb_path = osp.join(self.path, "bedroom_train_lmdb")
self.env = lmdb.open(lmdb_path, subdir=osp.isdir(lmdb_path),max_readers=1, readonly=True, lock=False,
readahead=False, meminit=False)
with self.env.begin(write=False) as txn:
self.length = txn.stat()["entries"]
root_split = lmdb_path.split("/")
cache_file = os.path.join("/".join(root_split[:-1]), f"_cache_{root_split[-1]}")
print('the cache file dir is {0}'.format(cache_file))
if os.path.isfile(cache_file):
self.keys = pickle.load(open(cache_file, "rb"))
else:
logging.info('Create the file to store')
with self.env.begin(write=False) as txn:
self.keys = []
ans = 0
for key,_ in txn.cursor():
self.keys.append(key)
pickle.dump(self.keys, open(cache_file, "wb"))
def __len__(self):
return len(self.keys)
def __getitem__(self, index):
img, target = None, torch.zeros(1)
env = self.env
with env.begin(write=False) as txn:
imgbuf = txn.get(self.keys[index])
buf = io.BytesIO()
buf.write(imgbuf)
buf.seek(0)
im = np.array(Image.open(buf).convert('RGB'))
image_size = 256
im = numpy.array(Image.fromarray(im).resize((image_size,image_size)))
#im = imresize(im, (image_size, image_size))
im = im / 256
im = im + np.random.uniform(0, 1 / 256., im.shape)
im = im.transpose((2, 0, 1))
im = (im-0.5)/0.5
#im_corrupt = np.random.uniform(0, 1, size=(image_size, image_size, 3))
return im, target
class CelebA(Dataset):
def __init__(self, cond_idx=1, filter_idx=0):
self.path = "/datasets01/CelebA/CelebA/072017/img_align_celeba/"
self.labels = pd.read_csv("/private/home/yilundu/list_attr_celeba.txt", sep="\s+", skiprows=1)
self.cond_idx = cond_idx
self.filter_idx = filter_idx
if filter_idx != 0:
mask = (self.labels.to_numpy()[:, self.cond_idx] == filter_idx)
self.labels = self.labels[mask].reset_index()
def __len__(self):
return self.labels.shape[0]
def __getitem__(self, index):
if FLAGS.single:
index = 0
info = self.labels.iloc[index]
if self.filter_idx != 0:
fname = info['index']
else:
fname = info.name
path = osp.join(self.path, fname)
im = imread(path)
im = imresize(im, (128, 128))
image_size = 128
im = im / 255.
label = int(info.iloc[self.cond_idx])
if label == -1:
label = 0
label = np.eye(2)[label]
if FLAGS.datasource == 'default':
im_corrupt = im + 0.3 * np.random.randn(image_size, image_size, 3)
elif FLAGS.datasource == 'random':
im_corrupt = np.random.uniform(
0, 1, size=(image_size, image_size, 3))
return im_corrupt, im, label
class Cifar10(Dataset):
def __init__(
self,
FLAGS,
train=True,
full=False,
augment=False,
noise=True,
rescale=1.0):
if augment:
transform_list = [
torchvision.transforms.RandomCrop(32, padding=4),
torchvision.transforms.RandomHorizontalFlip(),
torchvision.transforms.ToTensor(),
]
transform = transforms.Compose(transform_list)
else:
transform = transforms.ToTensor()
self.full = full
self.data = CIFAR10(
"data/cifar10",
transform=transform,
train=train,
download=True)
self.test_data = CIFAR10(
"data/cifar10",
transform=transform,
train=False,
download=True)
self.one_hot_map = np.eye(10)
self.noise = noise
self.rescale = rescale
self.FLAGS = FLAGS
def __len__(self):
if self.full:
return len(self.data) + len(self.test_data)
else:
return len(self.data)
def __getitem__(self, index):
FLAGS = self.FLAGS
if self.full:
if index >= len(self.data):
im, label = self.test_data[index - len(self.data)]
else:
im, label = self.data[index]
else:
im, label = self.data[index]
im = np.transpose(im, (1, 2, 0)).numpy()
image_size = 32
label = self.one_hot_map[label]
im = im * 255 / 256
im = im * self.rescale + \
np.random.uniform(0, 1 / 256., im.shape)
# np.random.seed((index + int(time.time() * 1e7)) % 2**32)
im_corrupt = np.random.uniform(
0.0, self.rescale, (image_size, image_size, 3))
return torch.Tensor(im_corrupt), torch.Tensor(im), label
class Cifar100(Dataset):
def __init__(self, FLAGS, train=True, augment=False):
transform = transforms.ToTensor()
self.one_hot_map = np.eye(100)
self.data = CIFAR100(
"/tmp/cifar100",
transform=transform,
train=train,
download=True)
def __len__(self):
return len(self.data)
def __getitem__(self, index):
im, label = self.data[0]
im = np.transpose(im, (1, 2, 0)).numpy()
image_size = 32
label = self.one_hot_map[label]
im = im * 255 / 256
im = im + \
np.random.uniform(0, 1 / 256., im.shape)
im_corrupt = np.random.uniform(
0.0, 1.0, (image_size, image_size, 3))
return im_corrupt, im, label
class Mnist(Dataset):
def __init__(self, train=True, rescale=1.0):
self.data = MNIST(
"data/mnist",
transform=transforms.ToTensor(),
download=True, train=train)
self.labels = np.eye(10)
def __len__(self):
return len(self.data)
def __getitem__(self, index):
im, label = self.data[index]
label = self.labels[label]
im = im.squeeze()
im = im.numpy() / 256 * 255 + np.random.uniform(0, 1. / 256, (28, 28))
im = np.clip(im, 0, 1)
s = 28
im_corrupt = np.random.uniform(0, 1, (s, s, 1))
im = im[:, :, None]
return torch.Tensor(im_corrupt), torch.Tensor(im), label
def infiniteloop(dataloader):
while True:
for x, y in iter(dataloader):
yield x
if __name__ == "__main__":
dataset = LSUNBed()
#data = dataset[100000]
print(len(dataset))
#img = data[0]
#print(img.shape)
#img = (img+1)/2
#image = Image.fromarray(np.uint8(img*255)).convert("RGB")
#image.save("bed.png")
#dataloader = torch.utils.data.DataLoader(
# dataset, 256, shuffle=True,
# num_workers=0, drop_last=True)
#datalooper = infiniteloop(dataloader)
#print(next(datalooper).size())
#data = pd.read_csv('sample/likelihood/ddpm1000.csv')
#print(data)
#lsun_data = LSUN('bedroom', 'train')
#torchvision.datasets.LSUN(classes='train', transform=None, target_transform=None)