-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathimage_processing.py
314 lines (263 loc) · 12.2 KB
/
image_processing.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
# Copyright 2016 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Read and preprocess image data.
Image processing occurs on a single image at a time. Image are read and
preprocessed in parallel across multiple threads. The resulting images
are concatenated together to form a single batch for training or evaluation.
-- Provide processed image data for a network:
inputs: Construct batches of evaluation examples of images.
distorted_inputs: Construct batches of training examples of images.
batch_inputs: Construct batches of training or evaluation examples of images.
-- Data processing:
parse_example_proto: Parses an Example proto containing a training example
of an image.
-- Image decoding:
decode_jpeg: Decode a JPEG encoded string into a 3-D float32 Tensor.
-- Image preprocessing:
image_preprocessing: Decode and preprocess one image for evaluation or training
distort_image: Distort one image for training a network.
eval_image: Prepare one image for evaluation.
distort_color: Distort the color in one image for training.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import tensorflow as tf
def decode_jpeg(image_buffer, scope=None):
"""Decode a JPEG string into one 3-D float image Tensor.
Args:
image_buffer: scalar string Tensor.
scope: Optional scope for name_scope.
Returns:
3-D float Tensor with values ranging from [0, 1).
"""
with tf.name_scope(values=[image_buffer], name=scope,
default_name='decode_jpeg'):
# Decode the string as an RGB JPEG.
# Note that the resulting image contains an unknown height and width
# that is set dynamically by decode_jpeg. In other words, the height
# and width of image is unknown at compile-time.
image = tf.image.decode_jpeg(image_buffer, channels=3)
# After this point, all image pixels reside in [0,1)
# until the very end, when they're rescaled to (-1, 1). The various
# adjust_* ops all require this range for dtype float.
image = tf.image.convert_image_dtype(image, dtype=tf.float32)
return image
def distort_color(image, thread_id=0, scope=None):
"""Distort the color of the image.
Each color distortion is non-commutative and thus ordering of the color ops
matters. Ideally we would randomly permute the ordering of the color ops.
Rather than adding that level of complication, we select a distinct ordering
of color ops for each preprocessing thread.
Args:
image: Tensor containing single image.
thread_id: preprocessing thread ID.
scope: Optional scope for name_scope.
Returns:
color-distorted image
"""
with tf.name_scope(values=[image], name=scope, default_name='distort_color'):
color_ordering = thread_id % 2
if color_ordering == 0:
image = tf.image.random_brightness(image, max_delta=32. / 255.)
image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
image = tf.image.random_hue(image, max_delta=0.2)
image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
elif color_ordering == 1:
image = tf.image.random_brightness(image, max_delta=32. / 255.)
image = tf.image.random_contrast(image, lower=0.5, upper=1.5)
image = tf.image.random_saturation(image, lower=0.5, upper=1.5)
image = tf.image.random_hue(image, max_delta=0.2)
# The random_* ops do not necessarily clamp.
image = tf.clip_by_value(image, 0.0, 1.0)
return image
def distort_image(image, height, width, bbox, thread_id=0, use_summary=False, scope=None):
"""Distort one image for training a network.
Distorting images provides a useful technique for augmenting the data
set during training in order to make the network invariant to aspects
of the image that do not effect the label.
Args:
image: 3-D float Tensor of image
height: integer
width: integer
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged
as [ymin, xmin, ymax, xmax].
thread_id: integer indicating the preprocessing thread.
scope: Optional scope for name_scope.
Returns:
3-D float Tensor of distorted image used for training.
"""
with tf.name_scope(values=[image, height, width, bbox], name=scope,
default_name='distort_image'):
# Each bounding box has shape [1, num_boxes, box coords] and
# the coordinates are ordered [ymin, xmin, ymax, xmax].
# Display the bounding box in the first thread only.
if use_summary:
image_with_box = tf.image.draw_bounding_boxes(tf.expand_dims(image, 0),
bbox)
tf.summary.image('image_with_bounding_boxes', image_with_box)
# A large fraction of image datasets contain a human-annotated bounding
# box delineating the region of the image containing the object of interest.
# We choose to create a new bounding box for the object which is a randomly
# distorted version of the human-annotated bounding box that obeys an allowed
# range of aspect ratios, sizes and overlap with the human-annotated
# bounding box. If no box is supplied, then we assume the bounding box is
# the entire image.
sample_distorted_bounding_box = tf.image.sample_distorted_bounding_box(
tf.shape(image),
bounding_boxes=bbox,
min_object_covered=0.1,
aspect_ratio_range=[0.75, 1.33],
area_range=[0.05, 1.0],
max_attempts=100,
use_image_if_no_bounding_boxes=True)
bbox_begin, bbox_size, distort_bbox = sample_distorted_bounding_box
if use_summary:
image_with_distorted_box = tf.image.draw_bounding_boxes(
tf.expand_dims(image, 0), distort_bbox)
tf.summary.image('images_with_distorted_bounding_box',
image_with_distorted_box)
# Crop the image to the specified bounding box.
distorted_image = tf.slice(image, bbox_begin, bbox_size)
# This resizing operation may distort the images because the aspect
# ratio is not respected. We select a resize method in a round robin
# fashion based on the thread number.
# Note that ResizeMethod contains 4 enumerated resizing methods.
resize_method = thread_id % 4
distorted_image = tf.image.resize_images(distorted_image, [height, width],
method=resize_method)
# Restore the shape since the dynamic slice based upon the bbox_size loses
# the third dimension.
distorted_image.set_shape([height, width, 3])
if use_summary:
tf.summary.image('cropped_resized_image',
tf.expand_dims(distorted_image, 0))
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
# Randomly distort the colors.
distorted_image = distort_color(distorted_image, thread_id)
if use_summary:
tf.summary.image('final_distorted_image',
tf.expand_dims(distorted_image, 0))
return distorted_image
def eval_image(image, height, width, scope=None):
"""Prepare one image for evaluation.
Args:
image: 3-D float Tensor
height: integer
width: integer
scope: Optional scope for name_scope.
Returns:
3-D float Tensor of prepared image.
"""
with tf.name_scope(values=[image, height, width], name=scope,
default_name='eval_image'):
# Crop the central region of the image with an area containing 87.5% of
# the original image.
image = tf.image.central_crop(image, central_fraction=0.875)
# Resize the image to the original height and width.
image = tf.expand_dims(image, 0)
image = tf.image.resize_bilinear(image, [height, width],
align_corners=False)
image = tf.squeeze(image, [0])
return image
def image_preprocessing(image_buffer, bbox, train, image_size, thread_id=0, use_summary=False):
"""Decode and preprocess one image for evaluation or training.
Args:
image_buffer: JPEG encoded string Tensor
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged as
[ymin, xmin, ymax, xmax].
train: boolean
thread_id: integer indicating preprocessing thread
Returns:
3-D float Tensor containing an appropriately scaled image
Raises:
ValueError: if user does not provide bounding box
"""
if bbox is None:
raise ValueError('Please supply a bounding box.')
image = decode_jpeg(image_buffer)
height = image_size
width = image_size
if train:
image = distort_image(image, height, width, bbox, thread_id=thread_id, use_summary=use_summary)
else:
image = eval_image(image, height, width)
# Finally, rescale to [-1,1] instead of [0, 1)
image = tf.subtract(image, 0.5)
image = tf.multiply(image, 2.0)
return image
def parse_example_proto(example_serialized):
"""Parses an Example proto containing a training example of an image.
The output of the build_image_data.py image preprocessing script is a dataset
containing serialized Example protocol buffers. Each Example proto contains
the following fields:
image/height: 462
image/width: 581
image/colorspace: 'RGB'
image/channels: 3
image/class/label: 615
image/class/synset: 'n03623198'
image/class/text: 'knee pad'
image/object/bbox/xmin: 0.1
image/object/bbox/xmax: 0.9
image/object/bbox/ymin: 0.2
image/object/bbox/ymax: 0.6
image/object/bbox/label: 615
image/format: 'JPEG'
image/filename: 'ILSVRC2012_val_00041207.JPEG'
image/encoded: <JPEG encoded string>
Args:
example_serialized: scalar Tensor tf.string containing a serialized
Example protocol buffer.
Returns:
image_buffer: Tensor tf.string containing the contents of a JPEG file.
label: Tensor tf.int32 containing the label.
bbox: 3-D float Tensor of bounding boxes arranged [1, num_boxes, coords]
where each coordinate is [0, 1) and the coordinates are arranged as
[ymin, xmin, ymax, xmax].
text: Tensor tf.string containing the human-readable label.
"""
# Dense features in Example proto.
feature_map = {
'image/encoded': tf.FixedLenFeature([], dtype=tf.string,
default_value=''),
'image/class/label': tf.FixedLenFeature([1], dtype=tf.int64,
default_value=-1),
'image/class/text': tf.FixedLenFeature([], dtype=tf.string,
default_value=''),
}
sparse_float32 = tf.VarLenFeature(dtype=tf.float32)
# Sparse features in Example proto.
feature_map.update(
{k: sparse_float32 for k in ['image/object/bbox/xmin',
'image/object/bbox/ymin',
'image/object/bbox/xmax',
'image/object/bbox/ymax']})
features = tf.parse_single_example(example_serialized, feature_map)
label = tf.cast(features['image/class/label'], dtype=tf.int32)
xmin = tf.expand_dims(features['image/object/bbox/xmin'].values, 0)
ymin = tf.expand_dims(features['image/object/bbox/ymin'].values, 0)
xmax = tf.expand_dims(features['image/object/bbox/xmax'].values, 0)
ymax = tf.expand_dims(features['image/object/bbox/ymax'].values, 0)
# Note that we impose an ordering of (y, x) just to make life difficult.
bbox = tf.concat(axis=0, values=[ymin, xmin, ymax, xmax])
# Force the variable number of bounding boxes into the shape
# [1, num_boxes, coords].
bbox = tf.expand_dims(bbox, 0)
bbox = tf.transpose(bbox, [0, 2, 1])
return features['image/encoded'], label, bbox, features['image/class/text']