-
Notifications
You must be signed in to change notification settings - Fork 0
/
miniimagenet_loader.py
371 lines (313 loc) · 12.2 KB
/
miniimagenet_loader.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
"""
Loading and using the Mini-ImageNet dataset.
To use these APIs, you should prepare a directory that
contains three sub-directories: train, test, and val.
Each of these three directories should contain one
sub-directory per WordNet ID.
"""
import os
import random
from collections import OrderedDict
import torch
from PIL import Image, ImageFile
from hbconfig import Config
from torch import nn
from torch.autograd import Variable
from torch.optim import Adam
from torchvision.transforms import transforms, ToTensor
from AutoAugment.autoaugment import ImageNetPolicy
from utils import find_latest, mkdir_p, get_sorted_path
ImageFile.LOAD_TRUNCATED_IMAGES = True
totensor = ToTensor()
def read_dataset_test(data_dir, transforms=None):
"""
Read the Mini-ImageNet dataset.
Args:
data_dir: directory containing Mini-ImageNet.
Returns:
A tuple (train, val, test) of sequences of
ImageNetClass instances.
"""
return tuple([_read_classes(os.path.join(data_dir, 'test'), transforms)])
def read_dataset(data_dir, transform_train=None, transform_test=None):
"""
Read the Mini-ImageNet dataset.
Args:
data_dir: directory containing Mini-ImageNet.
Returns:
A tuple (train, val, test) of sequences of
ImageNetClass instances.
"""
return tuple([_read_classes(os.path.join(data_dir, 'train'), transform_train), _read_classes(os.path.join(data_dir, 'val'), transform_test)]) # , 'test'
def _read_classes(dir_path, transforms):
"""
Read the WNID directories in a directory.
"""
print(f"=>Number of test classes = {len([ f for f in os.listdir(dir_path) if f.startswith('n')])}")
return [ImageNetClass(os.path.join(dir_path, f), transforms) for f in os.listdir(dir_path) if f.startswith('n')]
# pylint: disable=R0903
class ImageNetClass:
"""
A single image class.
"""
def __init__(self, dir_path, transform):
self.dir_path = dir_path
self._cache = {}
self.transform = transform
if transform is None:
self.transform = transforms.Compose([transforms.Resize((Config.data.image_size, Config.data.image_size)), # transforms.ToTensor(),
# transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])
def __len__(self):
return len([f for f in os.listdir(self.dir_path) if f.endswith('.JPEG')])
def sample(self, num_images):
"""
Sample images (as numpy arrays) from the class.
Returns:
A sequence of 84x84x3 numpy arrays.
Each pixel ranges from 0 to 1.
"""
names = [f for f in os.listdir(self.dir_path) if f.endswith('.JPEG')]
random.shuffle(names)
images = []
for name in names[:num_images]:
images.append(self._read_image(name))
return images
def _read_image(self, name):
if name in self._cache:
tmp = self._cache[name] # .astype('float32') / 0xff
return tmp
with open(os.path.join(self.dir_path, name), 'rb') as in_file:
img = Image.open(in_file).resize((Config.data.image_size, Config.data.image_size)).convert('RGB')
self._cache[name] = self.transform(img)
return self._read_image(name)
def _sample_mini_dataset(dataset, num_classes, num_shots):
"""
Sample a few shot task from a dataset.
Returns:
An iterable of (input, label) pairs.
"""
shuffled = list(dataset)
random.shuffle(shuffled)
for class_idx, class_obj in enumerate(shuffled[:num_classes]):
# num_shots = min(num_shots,len(class_obj))
for sample in class_obj.sample(num_shots):
yield (sample, class_idx)
def _mini_batches(samples, batch_size, num_batches, replacement):
"""
Generate mini-batches from some data.
Returns:
An iterable of sequences of (input, label) pairs,
where each sequence is a mini-batch.
"""
totensor = ToTensor()
samples = list(samples)
if replacement:
for _ in range(num_batches):
yield random.sample(totensor(samples), batch_size)
return
cur_batch = []
batch_count = 0
while True:
random.shuffle(samples)
for sample in samples:
cur_batch.append((totensor(sample[0]), sample[1]))
if len(cur_batch) < batch_size:
continue
yield cur_batch
cur_batch = []
batch_count += 1
if batch_count == num_batches:
return
class Interpolate(nn.Module):
def __init__(self, mode, scale_factor):
super(Interpolate, self).__init__()
self.interp = nn.functional.interpolate
self.size = scale_factor
self.mode = mode
def forward(self, x):
x = self.interp(x, size=self.size, mode=self.mode, align_corners=False)
return x
class GaussianNoise(nn.Module):
"""Gaussian noise regularizer.
Args:
sigma (float, optional): relative standard deviation used to generate the
noise. Relative means that it will be multiplied by the magnitude of
the value your are adding the noise to. This means that sigma can be
the same regardless of the scale of the vector.
is_relative_detach (bool, optional): whether to detach the variable before
computing the scale of the noise. If `False` then the scale of the noise
won't be seen as a constant but something to optimize: this will bias the
network to generate vectors with smaller values.
"""
def __init__(self, sigma=0.1, is_relative_detach=True):
super().__init__()
self.sigma = sigma
self.is_relative_detach = is_relative_detach
# self.noise = torch.FloatTensor(0)
def forward(self, x):
if self.training and self.sigma != 0:
scale = self.sigma * x.detach() if self.is_relative_detach else self.sigma * x
self.noise = torch.zeros(x.size())
sampled_noise = self.noise.normal_() * scale
x = x + sampled_noise
return x
class DynamicGNoise(nn.Module):
def __init__(self, shape, std=0.05):
super().__init__()
self.noise = Variable(torch.zeros(shape,shape).cuda())
self.std = std
def forward(self, x):
if not self.training: return x
self.noise.data.normal_(0, std=self.std)
print(x.size(), self.noise.size())
return x + self.noise
def gaussian(ins, is_training=True, mean=0.0, stddev=0.1):
if is_training:
noise = Variable(ins.data.new(ins.size()).normal_(mean, stddev))
return ins + noise
return ins
class AutoEncoder(nn.Module):
def __init__(self):
super(AutoEncoder, self).__init__()
self.path_to_save = Config.train.model_dir
self.use_cuda = torch.cuda.is_available()
# conv layers: (in_channel size, out_channels size, kernel_size, stride, padding)
# self.n_filters = 32
# self.layer1 = _conv_layer(3, self.n_filters, 3, 1, 0)
# self.layer2 = _conv_layer(self.n_filters, self.n_filters // 2, 3, 1, 0)
# self.layer3 = _conv_layer(self.n_filters // 2, self.n_filters // 4, 3, 1, 0)
# # self.emb = nn.Sequential(nn.Linear(512, 512))
# self.emb = nn.Sequential(nn.Linear(5408, 5408))
# # deconv layers: (in_channel size, out_channel size, kernel_size, stride, padding, output_padding)
# self.deconv1 = _conv_transpose_layer(self.n_filters // 4, self.n_filters // 2, 3, stride=3, padding=1, output_padding=1)
# self.deconv2 = _conv_transpose_layer(self.n_filters // 2, self.n_filters, 3, stride=2, padding=2, output_padding=1)
# self.deconv3 = _conv_transpose_layer(self.n_filters, 3, 3, stride=2, padding=3, output_padding=1)
self.encoder = nn.Sequential(nn.Conv2d(3, 32, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(True), nn.MaxPool2d(2),
nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(True), nn.MaxPool2d(2),
nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1), nn.LeakyReLU(True), nn.MaxPool2d(2), # nn.Linear(64*3*3,400),nn.LeakyReLU(True),
# nn.Linear(400,256),nn.LeakyReLU(True)
# nn.Conv2d(16, 8, kernel_size=3, padding=1), nn.LeakyReLU(True), nn.MaxPool2d(2),
#
# nn.Conv2d(8, 8, kernel_size=3, padding=1), nn.LeakyReLU(True), nn.MaxPool2d(2))
)
self.decoder = nn.Sequential( # nn.Linear(256,400), nn.LeakyReLU(True),
# nn.Linear(400,3*3*64), nn.LeakyReLU(True),
# Interpolate(mode='bilinear', scale_factor=2),
# GaussianNoise(sigma=0.1),
# DynamicGNoise(),
nn.ConvTranspose2d(64, 64, kernel_size=3, stride=2, padding=1, output_padding=1), nn.LeakyReLU(True),
# Interpolate(mode='bilinear', scale_factor=2),
nn.ConvTranspose2d(64, 32, kernel_size=3, stride=2, padding=1, output_padding=1), nn.LeakyReLU(True),
# Interpolate(mode='bilinear', scale_factor=2),
nn.ConvTranspose2d(32, 3, kernel_size=3, stride=2, padding=1, output_padding=1), nn.LeakyReLU(True),
# Interpolate(mode='bilinear', scale_factor=2),
# nn.ConvTranspose2d(32, 64, kernel_size=3,stride=2), nn.LeakyReLU(True),
# Interpolate(mode='bilinear', scale_factor=2),
# nn.ConvTranspose2d(64, 3, kernel_size=3,stride=2,padding=1), nn.LeakyReLU(True),
nn.Tanh())
mkdir_p(self.path_to_save)
# if self.use_cuda:
# self = self.cuda()
# summary(self.cuda(), (Config.data.channels, Config.data.image_size, Config.data.image_size))
# else:
# summary(self, (Config.data.channels, Config.data.image_size, Config.data.image_size))
def forward(self, x):
# print("Start Encode: ", x.shape)
if Config.train.add_noise:
x= gaussian(x,mean=Config.train.noise_mean,stddev=Config.train.noise_std)
x = self.encoder(x)
# x+= gaussian(0.2*x[0],mean=0.1,stddev=0.1)
# print("Finished Encode: ", x.shape)
x = self.decoder(x)
# print("Finished Decode: ", x.shape)
return x
def load_saved_model(self, path, model):
latest_path = find_latest(path + "/")
if latest_path is None:
return 0, model
checkpoint = torch.load(latest_path)
step_count = checkpoint['step_count']
state_dict = checkpoint['net']
# if dataparallel
# if "module" in list(state_dict.keys())[0]:
try:
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:] # remove 'module.' of dataparallel
new_state_dict[name] = v
model.load_state_dict(new_state_dict)
except:
# else:
model.load_state_dict(checkpoint['net'])
print(f"Load checkpoints...! {latest_path}")
return step_count, model
def set_optimizer(self):
self.optimizer = Adam(self.parameters(), lr=Config.train.learning_rate)
def save_checkpoint(self, step, max_to_keep=3):
sorted_path = get_sorted_path(self.path_to_save)
for i in range(len(sorted_path) - max_to_keep):
os.remove(sorted_path[i])
full_path = os.path.join(self.path_to_save, f"ae_{step}.pkl")
torch.save({"step_count": step, 'net': self.state_dict(), 'optimizer': self.optimizer.state_dict(), }, full_path)
print(f"Save checkpoints...! {full_path}")
def _mini_batches_with_augmentation(samples, batch_size, num_batches, replacement, num_aug=5, policy=None, use_cuda=False):
policy = policy # ImageNetPolicy()
if policy is None:
policy = ImageNetPolicy()
samples = list(samples)
cur_batch = []
if replacement:
for _ in range(num_batches):
for _ in range(num_aug):
for x in samples:
cur_batch.append((totensor(policy(x[0])), x[1]))
yield random.sample(cur_batch, batch_size)
return
batch_count = 0
while True:
random.shuffle(samples)
for idx in range(num_aug):
for sample in samples:
if idx == 0:
cur_batch.append((totensor(sample[0]), sample[1]))
else:
if use_cuda:
img = torch.unsqueeze(totensor(sample[0]), 0).cuda()
else:
img = torch.unsqueeze(totensor(sample[0]), 0)
if isinstance(policy,ImageNetPolicy):
cur_batch.append((totensor(policy(sample[0])).squeeze(), sample[1]))
else:
cur_batch.append((policy(img).squeeze(), sample[1]))
if len(cur_batch) < batch_size:
continue
yield cur_batch
cur_batch = []
batch_count += 1
if batch_count == num_batches:
return
def _split_train_test(samples, test_shots=1):
"""
Split a few-shot task into a train and a test set.
Args:
samples: an iterable of (input, label) pairs.
test_shots: the number of examples per class in the
test set.
Returns:
A tuple (train, test), where train and test are
sequences of (input, label) pairs.
"""
train_set = list(samples)
test_set = []
labels = set(item[1] for item in train_set)
for _ in range(test_shots):
for label in labels:
for i, item in enumerate(train_set):
if item[1] == label:
del train_set[i]
test_set.append(item)
break
if len(test_set) < len(labels) * test_shots:
raise IndexError('not enough examples of each class for test set')
return train_set, test_set