-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
39 lines (29 loc) · 1.28 KB
/
utils.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
import tensorflow as tf
import math
def save_model(root, checkpoint_prefix):
root.save(file_prefix=checkpoint_prefix)
def tf_record_parser(record):
keys_to_features = {
"image": tf.FixedLenFeature((), tf.string, default_value=""),
"height": tf.FixedLenFeature((), tf.int64),
"width": tf.FixedLenFeature((), tf.int64)
}
features = tf.parse_single_example(record, keys_to_features)
height = tf.cast(features['height'], tf.int64)
width = tf.cast(features['width'], tf.int64)
image = tf.decode_raw(features['image'], tf.uint8)
# reshape input and annotation images
image = tf.reshape(image, (height, width, 3), name="image_reshape")
image = tf.image.random_flip_left_right(image)
image = tf.image.resize_image_with_crop_or_pad(image, 128, 128)
# image = tf.contrib.image.rotate([image], tf.random_uniform(
# [1], maxval=math.pi / 10))[0]
return tf.to_float(image)
def normalizer(image, dtype):
# Not sure which one works better yet
image = tf.cast(image, dtype=dtype) / 255.0 - 0.5
# image = tf.cast(image, dtype=dtype) / 128.0 - 1.0
# noise addition normalization
image += tf.random_uniform(shape=tf.shape(image),
minval=0., maxval=1./128., dtype=dtype)
return image