-
Notifications
You must be signed in to change notification settings - Fork 1
/
bicubic_interp.py
124 lines (97 loc) · 3.49 KB
/
bicubic_interp.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
import tensorflow as tf
import numpy as np
def bicubic_interp_2d(input_, new_size):
"""
Args :
input_ : Input tensor. Its shape should be
[batch_size, height, width, channel].
In this implementation, the shape should be fixed for speed.
new_size : The output size [new_height, new_width]
ref : http://blog.demofox.org/2015/08/15/resizing-images-with-bicubic-interpolation/
"""
shape = input_.get_shape().as_list()
batch_size = shape[0]
height = shape[1]
width = shape[2]
channel = shape[3]
def _hermite(A, B, C, D, t):
a = A * -0.5 + B * 1.5 + C * -1.5 + D * 0.5
b = A + B * -2.5 + C * 2.0 + D * -0.5
c = A * -0.5 + C * 0.5
d = B
return a*t*t*t + b*t*t + c*t + d
def _get_grid_array(n_i, y_i, x_i, c_i):
n, y, x, c = np.meshgrid(n_i, y_i, x_i, c_i, indexing='ij')
n = np.expand_dims(n, axis=4)
y = np.expand_dims(y, axis=4)
x = np.expand_dims(x, axis=4)
c = np.expand_dims(c, axis=4)
return np.concatenate([n,y,x,c], axis=4)
def _get_frac_array(x_d, y_d, n, c):
x = x_d.shape[0]
y = y_d.shape[0]
x_t = x_d.reshape([1, 1, -1, 1])
y_t = y_d.reshape([1, -1, 1, 1])
y_t = np.tile(y_t, (n,1,x,c))
x_t = np.tile(x_t, (n,y,1,c))
return x_t, y_t
def _get_index_tensor(grid, x, y):
new_grid = np.array(grid)
grid_y = grid[:,:,:,:,1] + y
grid_x = grid[:,:,:,:,2] + x
grid_y = np.clip(grid_y, 0, height-1)
grid_x = np.clip(grid_x, 0, width-1)
new_grid[:,:,:,:,1] = grid_y
new_grid[:,:,:,:,2] = grid_x
return tf.constant(new_grid, dtype=tf.int32)
new_height = new_size[0]
new_width = new_size[1]
n_i = np.arange(batch_size)
c_i = np.arange(channel)
y_f = np.linspace(0., height-1, new_height)
y_i = y_f.astype(np.int32)
y_d = y_f - np.floor(y_f)
x_f = np.linspace(0., width-1, new_width)
x_i = x_f.astype(np.int32)
x_d = x_f - np.floor(x_f)
grid = _get_grid_array(n_i, y_i, x_i, c_i)
x_t, y_t = _get_frac_array(x_d, y_d, batch_size, channel)
i_00 = _get_index_tensor(grid, -1, -1)
i_10 = _get_index_tensor(grid, +0, -1)
i_20 = _get_index_tensor(grid, +1, -1)
i_30 = _get_index_tensor(grid, +2, -1)
i_01 = _get_index_tensor(grid, -1, +0)
i_11 = _get_index_tensor(grid, +0, +0)
i_21 = _get_index_tensor(grid, +1, +0)
i_31 = _get_index_tensor(grid, +2, +0)
i_02 = _get_index_tensor(grid, -1, +1)
i_12 = _get_index_tensor(grid, +0, +1)
i_22 = _get_index_tensor(grid, +1, +1)
i_32 = _get_index_tensor(grid, +2, +1)
i_03 = _get_index_tensor(grid, -1, +2)
i_13 = _get_index_tensor(grid, +0, +2)
i_23 = _get_index_tensor(grid, +1, +2)
i_33 = _get_index_tensor(grid, +2, +2)
p_00 = tf.gather_nd(input_, i_00)
p_10 = tf.gather_nd(input_, i_10)
p_20 = tf.gather_nd(input_, i_20)
p_30 = tf.gather_nd(input_, i_30)
p_01 = tf.gather_nd(input_, i_01)
p_11 = tf.gather_nd(input_, i_11)
p_21 = tf.gather_nd(input_, i_21)
p_31 = tf.gather_nd(input_, i_31)
p_02 = tf.gather_nd(input_, i_02)
p_12 = tf.gather_nd(input_, i_12)
p_22 = tf.gather_nd(input_, i_22)
p_32 = tf.gather_nd(input_, i_32)
p_03 = tf.gather_nd(input_, i_03)
p_13 = tf.gather_nd(input_, i_13)
p_23 = tf.gather_nd(input_, i_23)
p_33 = tf.gather_nd(input_, i_33)
col0 = _hermite(p_00, p_10, p_20, p_30, x_t)
col1 = _hermite(p_01, p_11, p_21, p_31, x_t)
col2 = _hermite(p_02, p_12, p_22, p_32, x_t)
col3 = _hermite(p_03, p_13, p_23, p_33, x_t)
value = _hermite(col0, col1, col2, col3, y_t)
return value
# Future : bicubic_interp_3d