1
|
|
|
# coding=utf-8 |
2
|
|
|
|
3
|
|
|
""" |
4
|
|
|
Tests for deepreg/model/backbone/global_net.py |
5
|
|
|
""" |
6
|
|
|
from typing import Tuple |
7
|
|
|
|
8
|
|
|
import pytest |
9
|
|
|
import tensorflow as tf |
10
|
|
|
|
11
|
|
|
from deepreg.model.backbone.global_net import AffineHead, GlobalNet |
12
|
|
|
|
13
|
|
|
|
14
|
|
|
def test_affine_head(): |
15
|
|
|
""" |
16
|
|
|
Test AffineHead. |
17
|
|
|
""" |
18
|
|
|
batch = 3 |
19
|
|
|
input_shape = (4, 5, 6) |
20
|
|
|
config = dict(image_size=input_shape, name="TestAffineHead") |
21
|
|
|
layer = AffineHead(**config) |
22
|
|
|
inputs = tf.ones(shape=(batch, *input_shape, 2)) |
23
|
|
|
ddf, theta = layer.call(inputs) |
24
|
|
|
assert ddf.shape == (batch, *input_shape, 3) |
25
|
|
|
assert theta.shape == (batch, 4, 3) |
26
|
|
|
|
27
|
|
|
got = layer.get_config() |
28
|
|
|
assert got == {"trainable": True, "dtype": "float32", **config} |
29
|
|
|
|
30
|
|
|
|
31
|
|
|
class TestGlobalNet: |
32
|
|
|
""" |
33
|
|
|
Test the backbone.global_net.GlobalNet class |
34
|
|
|
""" |
35
|
|
|
|
36
|
|
|
@pytest.mark.parametrize( |
37
|
|
|
"image_size,extract_levels,depth", |
38
|
|
|
[ |
39
|
|
|
((11, 12, 13), (0, 1, 2, 4), 4), |
40
|
|
|
((11, 12, 13), None, 4), |
41
|
|
|
((11, 12, 13), (0, 1, 2, 4), None), |
42
|
|
|
((8, 8, 8), (0, 1, 2), 3), |
43
|
|
|
], |
44
|
|
|
) |
45
|
|
|
def test_call( |
46
|
|
|
self, |
47
|
|
|
image_size: tuple, |
48
|
|
|
extract_levels: Tuple[int, ...], |
49
|
|
|
depth: int, |
50
|
|
|
): |
51
|
|
|
""" |
52
|
|
|
|
53
|
|
|
:param image_size: (dim1, dim2, dim3), dims of input image. |
54
|
|
|
:param extract_levels: from which depths the output will be built. |
55
|
|
|
:param depth: input is at level 0, bottom is at level depth |
56
|
|
|
""" |
57
|
|
|
batch_size = 5 |
58
|
|
|
out_ch = 3 |
59
|
|
|
network = GlobalNet( |
60
|
|
|
image_size=image_size, |
61
|
|
|
num_channel_initial=2, |
62
|
|
|
extract_levels=extract_levels, |
63
|
|
|
depth=depth, |
64
|
|
|
out_kernel_initializer="he_normal", |
65
|
|
|
out_activation="softmax", |
66
|
|
|
out_channels=out_ch, |
67
|
|
|
) |
68
|
|
|
inputs = tf.ones(shape=(batch_size, *image_size, out_ch)) |
69
|
|
|
ddf, theta = network.call(inputs) |
70
|
|
|
assert ddf.shape == inputs.shape |
71
|
|
|
assert theta.shape == (batch_size, 4, 3) |
72
|
|
|
|
73
|
|
|
def test_err(self): |
|
|
|
|
74
|
|
|
with pytest.raises(ValueError) as err_info: |
75
|
|
|
GlobalNet( |
76
|
|
|
image_size=(4, 5, 6), |
77
|
|
|
out_channels=3, |
78
|
|
|
num_channel_initial=2, |
79
|
|
|
depth=None, |
80
|
|
|
extract_levels=None, |
81
|
|
|
out_kernel_initializer="he_normal", |
82
|
|
|
out_activation="softmax", |
83
|
|
|
pooling=False, |
84
|
|
|
concat_skip=False, |
85
|
|
|
encode_kernel_sizes=[7, 3, 3], |
86
|
|
|
decode_kernel_sizes=3, |
87
|
|
|
strides=2, |
88
|
|
|
padding="same", |
89
|
|
|
name="Test", |
90
|
|
|
) |
91
|
|
|
assert "GlobalNet requires `depth` or `extract_levels`" in str(err_info.value) |
92
|
|
|
|
93
|
|
View Code Duplication |
def test_get_config(self): |
|
|
|
|
94
|
|
|
config = dict( |
95
|
|
|
image_size=(4, 5, 6), |
96
|
|
|
out_channels=3, |
97
|
|
|
num_channel_initial=2, |
98
|
|
|
depth=2, |
99
|
|
|
extract_levels=(2,), |
100
|
|
|
out_kernel_initializer="he_normal", |
101
|
|
|
out_activation="softmax", |
102
|
|
|
pooling=False, |
103
|
|
|
concat_skip=False, |
104
|
|
|
encode_kernel_sizes=[7, 3, 3], |
105
|
|
|
decode_kernel_sizes=3, |
106
|
|
|
encode_num_channels=[2, 4, 8], |
107
|
|
|
decode_num_channels=[2, 4, 8], |
108
|
|
|
strides=2, |
109
|
|
|
padding="same", |
110
|
|
|
name="Test", |
111
|
|
|
) |
112
|
|
|
network = GlobalNet(**config) |
113
|
|
|
got = network.get_config() |
114
|
|
|
assert got == config |
115
|
|
|
|