-
Notifications
You must be signed in to change notification settings - Fork 15
/
dataset.py
319 lines (266 loc) · 11.6 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
from __future__ import division
from __future__ import print_function
import numpy as np
import tensorflow as tf
def process_image(img, crop=None, mean=None,
mirror=True, is_training=True):
"""
Preprocessing code. For training this function randomly crop images and
flips the image randomly.
For testing we use the center crop of the image.
Args:
img: The input image.
crop: Size of the output image.
mean: three dimensional array indicating the mean values to subtract
from the image.
mirror: Flag, which indicates if images should be mirrored.
is_training: Flag which indicates whether training preprocessing
or testing preprocessing should be used.
Returns:
A pre-processed image.
"""
if is_training:
img = tf.random_crop(img, [crop, crop, img.get_shape().as_list()[
2]], name='random_image_crop')
if mirror:
img = tf.image.random_flip_left_right(img)
else:
new_shape = img.get_shape().as_list()[0]
offset = (new_shape - crop) // 2
img = tf.slice(img, begin=tf.convert_to_tensor(
[offset, offset, 0]), size=tf.convert_to_tensor([crop, crop, -1]))
# Mean subtraction
return tf.to_float(img) - mean
class NpyDatasetProvider(object):
"""
This class hooks up a numpy dataset file to tensorflow queue runners.
"""
def __init__(self, data_spec, image_file, label_file,
labels_per_batch=6, images_per_identity=None,
batch_size=32, is_training=True, num_concurrent=4):
super(NpyDatasetProvider, self).__init__()
# The data specifications describe how to process the image
self.data_spec = data_spec
self.images = np.transpose(np.load(image_file), (0, 2, 3, 1))
self.labels = np.load(label_file)
if self.labels.dtype != np.int32:
self.labels = self.labels.astype(np.int32)
self.original_num_images = len(self.images)
self.batch_size = batch_size
self.is_training = is_training
self.num_images = len(self.images)
self.images_per_identity = images_per_identity
if images_per_identity is not None:
batch_size = images_per_identity * labels_per_batch
self.batch_size = batch_size
if not self.is_training and self.num_images % self.batch_size != 0:
to_pad = self.batch_size - (self.num_images % self.batch_size)
pad_img = np.zeros(
[to_pad] + list(self.images.shape[1:]), dtype=np.uint8)
pad_label = -np.ones([to_pad], dtype=np.int32)
self.labels = np.r_[self.labels, pad_label]
self.images = np.vstack([self.images, pad_img])
self.num_images = len(self.images)
self.indices = np.arange(len(self.images))
self.labels_per_batch = labels_per_batch
self.unique_labels = np.unique(self.labels)
self.tf_unique_labels = tf.convert_to_tensor(self.unique_labels)
self.setup(num_concurrent)
def _setup_test(self, num_concurrent):
"""
Setup the test queue.
Args:
num_concurrent: Number of concurrent threads.
"""
num_images = len(self.images)
self.num_batches = num_images // self.batch_size
indices = tf.range(num_images)
self.preprocessing_queue = tf.FIFOQueue(capacity=len(self.images),
dtypes=[tf.int32],
shapes=[()],
name='preprocessing_queue')
self.test_queue_op = self.preprocessing_queue.enqueue_many([indices])
image_shape = (self.data_spec.crop_size,
self.data_spec.crop_size, self.data_spec.channels)
processed_queue = tf.FIFOQueue(capacity=len(self.images),
dtypes=[tf.int32, tf.float32],
shapes=[(), image_shape],
name='processed_queue')
label, img = self.process_test()
enqueue_processed_op = processed_queue.enqueue([label, img])
self.dequeue_op = processed_queue.dequeue_many(self.batch_size)
num_concurrent = min(num_concurrent, num_images)
self.queue_runner = tf.train.QueueRunner(
processed_queue,
[enqueue_processed_op] * num_concurrent)
tf.train.add_queue_runner(self.queue_runner)
def setup(self, num_concurrent):
"""
Setup of the queues.
Args:
num_concurrent: Number of concurrent threads.
"""
if self.is_training:
return self._setup_train(num_concurrent)
else:
return self._setup_test(num_concurrent)
def _setup_train(self, num_concurrent):
"""
Setup of the training queues.
Args:
num_concurrent: Number of concurrent threads.
"""
num_images = len(self.images)
self.num_batches = num_images // self.batch_size
# Crate a label queue.
self.label_queue = tf.RandomShuffleQueue(
capacity=len(self.unique_labels),
min_after_dequeue=self.labels_per_batch,
dtypes=[tf.int32],
shapes=[()],
name='label_queue')
self.label_queue_op = self.label_queue.enqueue_many(
[self.tf_unique_labels])
(labels, processed_images) = self.process()
# print(labels.get_shape().as_list())
# print(processed_images.get_shape().as_list())
image_shape = (self.data_spec.crop_size,
self.data_spec.crop_size, self.data_spec.channels)
processed_queue = tf.FIFOQueue( # capacity=len(self.images),
capacity=self.batch_size * 6,
dtypes=[tf.int32, tf.float32],
shapes=[(), image_shape],
name='processed_queue')
# Enqueue the processed image and path
enqueue_processed_op = processed_queue.enqueue_many(
[labels, processed_images])
self.dequeue_op = processed_queue.dequeue_many(self.batch_size)
num_concurrent = min(num_concurrent, num_images)
self.label_runner = tf.train.QueueRunner(
self.label_queue, [self.label_queue_op] * (num_concurrent + 1))
self.queue_runner = tf.train.QueueRunner(
processed_queue,
[enqueue_processed_op] * num_concurrent)
tf.train.add_queue_runner(self.label_runner)
tf.train.add_queue_runner(self.queue_runner)
def start(self, session, coordinator, num_concurrent=4):
"""
Start the processing worker threads.
Args:
session: A tensorflow session.
coordinator: A tensorflow coordinator.
num_concurrent: Number of concurrent threads.
Returns:
a create threads operation.
"""
if self.is_training:
self.label_runner.create_threads(
session, coord=coordinator, start=True)
else:
session.run(self.test_queue_op) # just enqueue labels once!
return self.queue_runner.create_threads(
session, coord=coordinator, start=True)
def feed_data(self, session):
"""
Call this function for testing NpyDatasetProvider. It pushes
the testing dataset once into the queue.
Args:
session: A tensorflow session
"""
assert(not self.is_training)
session.run(self.test_queue_op) # just enqueue labels once!
def get_labels(self, session):
"""
Returns a list of labels from the queue.
Args:
session: A tensorflow session.
Returns:
An array of labels from the queue.
"""
labels = session.run(
self.label_queue.dequeue_many(self.labels_per_batch))
return labels
def get(self, session):
"""
Get a single batch of images along with their labels.
Returns:
a tuple of (labels, images)
"""
(labels, images) = session.run(self.dequeue_op)
return (labels, images)
def batches(self, session):
"""
Yield a batch until no more images are left.
Yields:
Tuples in the form (labels, images)
"""
for _ in xrange(self.num_batches):
yield self.get(session=session)
def process_test(self):
"""
Processes the test images.
Returns:
Tuple consisting of (label, processed_image).
"""
def fetch_images(the_idx):
return self.images[the_idx, ...]
def fetch_labels(the_idx):
return self.labels[the_idx]
index = self.preprocessing_queue.dequeue()
label = tf.py_func(fetch_labels, [index], tf.int32)
label.set_shape([])
the_img = tf.py_func(fetch_images, [index], tf.uint8)
the_img.set_shape(self.images.shape[1:])
processed_img = process_image(img=the_img,
#img=self.tf_images[index, ...],
crop=self.data_spec.crop_size,
mean=self.data_spec.mean,
is_training=self.is_training)
# return (self.tf_labels[index], processed_img)
return (label, processed_img)
def process(self):
"""
Processes a training image.
Returns:
A tuple consisting of (label, image).
"""
# Dequeue a single image path
def fetch_data(sampled_labels):
if self.images_per_identity is not None:
all_ids = []
for label in sampled_labels:
valid_mask, = np.nonzero(self.labels == label)
try:
valid_mask = np.random.choice(
valid_mask, size=self.images_per_identity,
replace=False)
except:
valid_mask = np.random.choice(
valid_mask, size=self.images_per_identity,
replace=True) # well, whatever...
all_ids.append(valid_mask)
all_ids = np.concatenate(all_ids)
valid_labels = self.labels[all_ids]
valid_images = self.images[all_ids, ...]
return valid_labels, valid_images
else:
valid_mask, = np.nonzero(np.in1d(self.labels, sampled_labels))
valid_mask = np.random.choice(valid_mask, size=self.batch_size)
valid_labels = self.labels[valid_mask]
valid_images = self.images[valid_mask, ...]
return valid_labels, valid_images
labels = self.label_queue.dequeue_many(self.labels_per_batch)
labels.set_shape([self.labels_per_batch])
labels, images = tf.py_func(fetch_data, [labels], [tf.int32, tf.uint8])
labels.set_shape([self.batch_size])
images.set_shape([self.batch_size] + list(self.images.shape[1:]))
processed_images = []
for i in xrange(self.batch_size):
# Process the image
processed_img = process_image(img=images[i, ...],
crop=self.data_spec.crop_size,
mean=self.data_spec.mean,
is_training=self.is_training)
processed_images.append(processed_img)
processed_images = tf.stack(processed_images)
return (labels, processed_images)