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
| def transform_perspective(image):
def x_y_1():
x = tf.random_uniform([], minval=-0.3, maxval=-0.15)
y = tf.random_uniform([], minval=-0.3, maxval=-0.15)
return x, y
def x_y_2():
x = tf.random_uniform([], minval=0.15, maxval=0.3)
y = tf.random_uniform([], minval=0.15, maxval=0.3)
return x, y
def trans(image):
ran = tf.random_uniform([])
x = tf.random_uniform([], minval=-0.3, maxval=0.3)
x_com = tf.random_uniform([], minval=1-x-0.1, maxval=1-x+0.1)
y = tf.random_uniform([], minval=-0.3, maxval=0.3)
y_com = tf.random_uniform([], minval=1-y-0.1, maxval=1-y+0.1)
transforms = [x_com, x,0,y,y_com,0,0.00,0]
ran = tf.random_uniform([])
image = tf.cond(ran<0.5, lambda:tf.contrib.image.transform(image,transforms,interpolation='NEAREST', name=None),
lambda:tf.contrib.image.transform(image,transforms,interpolation='BILINEAR', name=None))
return image
ran = tf.random_uniform([])
image = tf.cond(ran<1, lambda: trans(image), lambda:image)
return image
|
