-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpre_process.py
147 lines (130 loc) · 6.72 KB
/
pre_process.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
import tensorflow as tf
import numpy as np
import math
def Gaussian(inputs, in_channels=3, dimensions=3,theta=0.8, multi_scale=False,name='gaussian_filter'):
'''
:param inputs:
:param name:
:return:
'''
if multi_scale:
Gaussian_kernel1 = Gaussian_template(theta=theta, accept_field=1, dimensions=dimensions)
Gaussian_kernel2 = Gaussian_template(theta=0.5, accept_field=1, dimensions=dimensions)
Gaussian_kernel3 = Gaussian_template(theta=0.3, accept_field=1, dimensions=dimensions)
size=Gaussian_kernel1.shape
dims=len(size)
size = (1,)+size if dims==2 else size
if in_channels==1:
pass
else:
Gaussian_kernel1 = np.expand_dims(Gaussian_kernel1,axis=dims)
Gaussian_kernel2 = np.expand_dims(Gaussian_kernel2,axis=dims)
Gaussian_kernel3 = np.expand_dims(Gaussian_kernel3,axis=dims)
_kernel1 = np.concatenate([Gaussian_kernel1 for _ in range(in_channels)], axis=dims)
_kernel2 = np.concatenate([Gaussian_kernel2 for _ in range(in_channels)], axis=dims)
_kernel3 = np.concatenate([Gaussian_kernel3 for _ in range(in_channels)], axis=dims)
Gaussian1 = tf.get_variable(name=name+'_1', shape=size + (in_channels, 1), dtype=tf.float32,
initializer=tf.constant_initializer(_kernel1),trainable=False)
Gaussian2 = tf.get_variable(name=name+'_2', shape=size + (in_channels, 1), dtype=tf.float32,
initializer=tf.constant_initializer(_kernel2),trainable=False)
Gaussian3 = tf.get_variable(name=name+'_3', shape=size + (in_channels, 1), dtype=tf.float32,
initializer=tf.constant_initializer(_kernel3),trainable=False)
filter_gaussian1 = tf.nn.conv3d(inputs, Gaussian1, (1, 1, 1, 1, 1), padding='SAME')
filter_gaussian2 = tf.nn.conv3d(inputs, Gaussian2, (1, 1, 1, 1, 1), padding='SAME')
filter_gaussian3 = tf.nn.conv3d(inputs, Gaussian3, (1, 1, 1, 1, 1), padding='SAME')
filter_gaussian = tf.concat([filter_gaussian1, filter_gaussian2, filter_gaussian3], axis=-1)
else:
Gaussian_kernel = Gaussian_template(theta=theta, accept_field=1, dimensions=dimensions)
size=Gaussian_kernel.shape
dims=len(size)
size = (1,)+size if dims==2 else size
Gaussian = tf.get_variable(name=name, shape=size + (in_channels, 1), dtype=tf.float32,
initializer=tf.constant_initializer(Gaussian_kernel),trainable=False)
filter_gaussian = tf.nn.conv3d(inputs, Gaussian, (1, 1, 1, 1, 1), padding='SAME')
return filter_gaussian
def Batch_Norm(inputs,name='NORM'):
with tf.variable_scope(name):
mean, variance = tf.nn.moments(inputs, axes=(0, 1, 2, 3), keep_dims=True, name="normalize_moments")
Gamma = tf.constant(1.0, name="scale_factor", shape=mean.shape, dtype=tf.float32)
Beta = tf.constant(0.0, name="offset_factor", shape=mean.shape, dtype=tf.float32)
_inputs = tf.nn.batch_normalization(inputs, mean, variance, offset=Beta, scale=Gamma, variance_epsilon=1e-3)
return _inputs
def get_temporal_kernel(size, in_channel, out_channel):
if size == 3:
laplace3 = np.array([[[[[1]]]], [[[[-2]]]], [[[[1]]]]], dtype=np.float32)
_kernel3 = np.concatenate([laplace3 for _ in range(out_channel)], axis=4)
if in_channel==1:
kernel3=_kernel3
else:
kernel3 = np.concatenate([_kernel3 for _ in range(in_channel)], axis=3)
return kernel3
elif size == 5:
laplace5 = np.array(
[[[[[0.25]]]], [[[[0]]]], [[[[-0.5]]]], [[[[0]]]],
[[[[0.25]]]]], dtype=np.float32)
_kernel5 = np.concatenate([laplace5 for _ in range(out_channel)], axis=4)
if in_channel==1:
kernel5=_kernel5
else:
kernel5 = np.concatenate([_kernel5 for _ in range(in_channel)], axis=3)
return kernel5
elif size == 7:
laplace7 = np.array(
[[[[[1 / 9]]]], [[[[0]]]], [[[[0]]]], [[[[-2 / 9]]]],
[[[[0]]]], [[[[0]]]], [[[[1 / 9]]]]], dtype=np.float32)
_kernel7 = np.concatenate([laplace7 for _ in range(out_channel)], axis=4)
if in_channel==1:
kernel7 = _kernel7
else:
kernel7 = np.concatenate([_kernel7 for _ in range(in_channel)], axis=3)
return kernel7
else:
assert False, 'Not supported size, must be 3/5/7'
def spatial_gradients(in_channel=3,mode='sobel', name='gaussian_filter'):
if mode=='sobel':
y_kernel = np.array(
[[[[[-1]],[[0]],[[1]]],[[[-2]],[[0]],[[2]]],[[[-1]],[[0]],[[1]]]]], dtype=np.float32) # (D,H,W,in_channel,out_channel)
x_kernel = np.array(
[[[[[-1]],[[-2]],[[-1]]],[[[0]],[[0]],[[0]]],[[[1]],[[2]],[[1]]]]], dtype=np.float32)
if in_channel==1:
return (y_kernel,x_kernel)
else:
g_y = np.concatenate([y_kernel for _ in range(in_channel)], axis=3)
g_x = np.concatenate([x_kernel for _ in range(in_channel)], axis=3)
return (g_y,g_x)
if mode=='roberts':
y_kernel = np.array(
[[[[[1]],[[0]]],[[[0]],[[-1]]]]], dtype=np.float32)
x_kernel = np.array(
[[[[[0]],[[1]]],[[[-1]],[[0]]]]], dtype=np.float32)
if in_channel==1:
return (y_kernel,x_kernel)
else:
g_y = np.concatenate([y_kernel for _ in range(in_channel)], axis=3)
g_x = np.concatenate([x_kernel for _ in range(in_channel)], axis=3)
return (g_y,g_x)
def Gaussian_template(theta,accept_field=1,dimensions=2):
# 均值为0，方差为theta的离散高斯模板，theta越大，曲线越矮胖
size = []
[size.append(2*accept_field+1) for _ in range(dimensions)]
H = np.zeros(shape=size,dtype=np.float32)
index = [range(0, 2*accept_field+1) for _ in range(dimensions)]
if len(index) == 3:
for i in index[0]:
for j in index[1]:
for k in index[2]:
H[i,j,k] = (1.0/(2*math.pi*theta**2))*math.exp(
-(pow(i-accept_field,2)+pow(j-accept_field,2)+pow(k-accept_field,2))/(2*theta**2))
elif len(index) == 2:
for i in index[0]:
for j in index[1]:
H[i, j] = (1.0 / (2 * math.pi * theta ** 2)) * math.exp(
-(pow(i - accept_field, 2) + pow(j - accept_field, 2)) / (2 * theta ** 2))
else:
assert False, 'only 2/3-dimension kernel supported now,if ' \
'you want customize you own kernel,change cycle times bellow'
return H
if __name__ == '__main__':
_k = spatial_gradients()
print(_k[0].shape,_k[1].shape)
print(_k[0],_k[1])