-
Notifications
You must be signed in to change notification settings - Fork 20
/
Copy pathdataloader.py
150 lines (114 loc) · 6.68 KB
/
dataloader.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
import tensorflow as tf
from collections import namedtuple
dataloader_parameters = namedtuple('dataloader_parameters',
'patch_height, patch_width, '
'height, width, '
'batch_size, '
'num_threads, ')
def string_length_tf(t):
return tf.py_func(len, [t], [tf.int64])
class Dataloader(object):
def __init__(self, data_path_image, data_path_proxy, filenames_file, dataset, is_training,
test_single, image_path, post_process, params):
self.data_path_image = data_path_image
self.data_path_proxy = data_path_proxy
self.params = params
self.dataset = dataset
self.image_path = image_path
self.is_training = is_training
self.test_single=test_single
self.post_process = post_process
self.left_image_batch = None
self.right_image_batch = None
self.proxy_left_batch = None
self.proxy_right_batch = None
if not self.test_single:
input_queue = tf.train.string_input_producer([filenames_file], shuffle=False)
line_reader = tf.TextLineReader()
_, line = line_reader.read(input_queue)
self.split_line = tf.string_split([line]).values
if is_training:
self.left_image_path = tf.string_join([self.data_path_image, self.split_line[0]])
self.right_image_path = tf.string_join([self.data_path_image, self.split_line[1]])
self.proxy_left_path = tf.string_join([self.data_path_proxy, self.split_line[2]])
self.proxy_right_path = tf.string_join([self.data_path_proxy, self.split_line[3]])
left_image_o = self.read_image(self.left_image_path)
right_image_o = self.read_image(self.right_image_path)
proxy_left_o = self.read_proxy(self.proxy_left_path)
proxy_right_o = self.read_proxy(self.proxy_right_path)
do_flip = tf.random_uniform([], 0, 1)
left_image = tf.cond(do_flip > 0.5, lambda: tf.image.flip_left_right(right_image_o), lambda: left_image_o)
right_image = tf.cond(do_flip > 0.5, lambda: tf.image.flip_left_right(left_image_o), lambda: right_image_o)
proxy_left = tf.cond(do_flip > 0.5, lambda: tf.image.flip_left_right(proxy_right_o), lambda: proxy_left_o)
proxy_right = tf.cond(do_flip > 0.5, lambda: tf.image.flip_left_right(proxy_left_o), lambda: proxy_right_o)
# randomly augment images
do_augment = tf.random_uniform([], 0, 1)
left_image, right_image = tf.cond(do_augment > 0.5,
lambda: self.augment_image_pair(left_image, right_image),
lambda: (left_image, right_image))
left_image.set_shape([None, None, 3])
right_image.set_shape([None, None, 3])
proxy_left.set_shape([None, None, 1])
proxy_right.set_shape([None, None, 1])
crops=tf.random_crop(tf.concat([left_image, right_image, proxy_left, proxy_right], -1), [self.params.patch_height, self.params.patch_width, 8])
left_image, right_image, proxy_left, proxy_right = tf.split(crops, [3, 3, 1, 1], axis=2)
min_after_dequeue = 32
capacity = min_after_dequeue + self.params.batch_size
self.left_image_batch, self.right_image_batch, self.proxy_left_batch, self.proxy_right_batch = \
tf.train.shuffle_batch([left_image, right_image, proxy_left, proxy_right], self.params.batch_size, capacity,
min_after_dequeue, self.params.num_threads)
else:
self.left_image_path = tf.cast([self.image_path], tf.string)[0] if self.test_single \
else tf.string_join([self.data_path_image, self.split_line[0]])
left_image_o = self.read_image(self.left_image_path)
if self.post_process:
self.left_image_batch = tf.stack([left_image_o, tf.image.flip_left_right(left_image_o)], 0)
self.left_image_batch.set_shape([2, None, None, 3])
else:
self.left_image_batch = tf.stack([left_image_o], 0)
self.left_image_batch.set_shape([1, None, None, 3])
def augment_image_pair(self, left_image, right_image):
# randomly shift gamma
random_gamma = tf.random_uniform([], 0.8, 1.2)
left_image_aug = left_image ** random_gamma
right_image_aug = right_image ** random_gamma
# randomly shift brightness
random_brightness = tf.random_uniform([], 0.5, 2.0)
left_image_aug = left_image_aug * random_brightness
right_image_aug = right_image_aug * random_brightness
# randomly shift color
random_colors = tf.random_uniform([3], 0.8, 1.2)
white = tf.ones([tf.shape(left_image)[0], tf.shape(left_image)[1]])
color_image = tf.stack([white * random_colors[i] for i in range(3)], axis=2)
left_image_aug *= color_image
right_image_aug *= color_image
# saturate
left_image_aug = tf.clip_by_value(left_image_aug, 0, 1)
right_image_aug = tf.clip_by_value(right_image_aug, 0, 1)
return left_image_aug, right_image_aug
def read_image(self, image_path):
path_length = string_length_tf(image_path)[0]
file_extension = tf.substr(image_path, path_length - 3, 3)
file_cond = tf.equal(file_extension, 'jpg')
image = tf.cond(file_cond, lambda: tf.image.decode_jpeg(tf.read_file(image_path)),
lambda: tf.image.decode_png(tf.read_file(image_path)))
if self.dataset == 'cityscapes':
o_height = tf.shape(image)[0]
crop_height = (o_height * 4) / 5
image = image[:crop_height, :, :]
self.image_w = tf.cast(tf.shape(image)[1], tf.float32)
self.image_h = tf.cast(tf.shape(image)[0], tf.float32)
image = tf.image.convert_image_dtype(image, tf.float32)
image = tf.image.resize_images(image, [self.params.height, self.params.width], tf.image.ResizeMethod.NEAREST_NEIGHBOR)
return image
def read_proxy(self, image_path):
image = tf.image.decode_png(tf.read_file(image_path), dtype=tf.uint16)
image = tf.cast(image, tf.float32)
if self.dataset == 'cityscapes':
o_height = tf.shape(image)[0]
crop_height = (o_height * 4) /5
image = image[:crop_height, :, :]
image_w = tf.cast(tf.shape(image)[1], tf.float32)
image = tf.image.resize_images(image, [self.params.height, self.params.width], tf.image.ResizeMethod.NEAREST_NEIGHBOR)/256.0
image = image * (self.params.width/image_w)
return image