1
|
|
|
from torchio.transforms import RandomLabelsToImage |
2
|
|
|
from ...utils import TorchioTestCase |
3
|
|
|
|
4
|
|
|
|
5
|
|
|
class TestRandomLabelsToImage(TorchioTestCase): |
6
|
|
|
"""Tests for `RandomLabelsToImage`.""" |
7
|
|
|
def test_random_simulation(self): |
8
|
|
|
"""The transform runs without error and an 'image_from_labels' key is |
9
|
|
|
present in the transformed subject.""" |
10
|
|
|
transform = RandomLabelsToImage(label_key='label') |
11
|
|
|
transformed = transform(self.sample_subject) |
12
|
|
|
self.assertIn('image_from_labels', transformed) |
13
|
|
|
|
14
|
|
|
def test_deterministic_simulation(self): |
15
|
|
|
"""The transform creates an image where values are equal to given |
16
|
|
|
mean if standard deviation is zero. |
17
|
|
|
Using a label map.""" |
18
|
|
|
transform = RandomLabelsToImage( |
19
|
|
|
label_key='label', |
20
|
|
|
mean=[0.5, 2], |
21
|
|
|
std=[0, 0] |
22
|
|
|
) |
23
|
|
|
transformed = transform(self.sample_subject) |
24
|
|
|
self.assertTensorEqual( |
25
|
|
|
transformed['image_from_labels'].data == 0.5, |
26
|
|
|
self.sample_subject['label'].data == 0 |
27
|
|
|
) |
28
|
|
|
self.assertTensorEqual( |
29
|
|
|
transformed['image_from_labels'].data == 2, |
30
|
|
|
self.sample_subject['label'].data == 1 |
31
|
|
|
) |
32
|
|
|
|
33
|
|
|
def test_deterministic_simulation_with_discretized_label_map(self): |
34
|
|
|
"""The transform creates an image where values are equal to given mean |
35
|
|
|
if standard deviation is zero. |
36
|
|
|
Using a discretized label map.""" |
37
|
|
|
transform = RandomLabelsToImage( |
38
|
|
|
label_key='label', |
39
|
|
|
mean=[0.5, 2], |
40
|
|
|
std=[0, 0], |
41
|
|
|
discretize=True |
42
|
|
|
) |
43
|
|
|
transformed = transform(self.sample_subject) |
44
|
|
|
self.assertTensorEqual( |
45
|
|
|
transformed['image_from_labels'].data == 0.5, |
46
|
|
|
self.sample_subject['label'].data == 0 |
47
|
|
|
) |
48
|
|
|
self.assertTensorEqual( |
49
|
|
|
transformed['image_from_labels'].data == 2, |
50
|
|
|
self.sample_subject['label'].data == 1 |
51
|
|
|
) |
52
|
|
|
|
53
|
|
|
def test_deterministic_simulation_with_pv_map(self): |
54
|
|
|
"""The transform creates an image where values are equal to given |
55
|
|
|
mean weighted by partial-volume if standard deviation is zero.""" |
56
|
|
|
subject = self.get_subject_with_partial_volume_label_map(components=2) |
57
|
|
|
transform = RandomLabelsToImage( |
58
|
|
|
label_key='label', |
59
|
|
|
mean=[0.5, 1], |
60
|
|
|
std=[0, 0] |
61
|
|
|
) |
62
|
|
|
transformed = transform(subject) |
63
|
|
|
self.assertTensorAlmostEqual( |
64
|
|
|
transformed['image_from_labels'].data[0], |
65
|
|
|
subject['label'].data[0] * 0.5 + subject['label'].data[1] * 1 |
66
|
|
|
) |
67
|
|
|
self.assertEqual( |
68
|
|
|
transformed['image_from_labels'].data.shape, |
69
|
|
|
(1, 10, 20, 30) |
70
|
|
|
) |
71
|
|
|
|
72
|
|
|
def test_deterministic_simulation_with_discretized_pv_map(self): |
73
|
|
|
"""The transform creates an image where values are equal to given mean |
74
|
|
|
if standard deviation is zero. |
75
|
|
|
Using a discretized partial-volume label map.""" |
76
|
|
|
subject = self.get_subject_with_partial_volume_label_map() |
77
|
|
|
transform = RandomLabelsToImage( |
78
|
|
|
label_key='label', |
79
|
|
|
mean=[0.5], |
80
|
|
|
std=[0], |
81
|
|
|
discretize=True |
82
|
|
|
) |
83
|
|
|
transformed = transform(subject) |
84
|
|
|
self.assertTensorAlmostEqual( |
85
|
|
|
transformed['image_from_labels'].data, |
86
|
|
|
(subject['label'].data > 0) * 0.5 |
87
|
|
|
) |
88
|
|
|
|
89
|
|
|
def test_filling(self): |
90
|
|
|
"""The transform can fill in the generated image with an already |
91
|
|
|
existing image. |
92
|
|
|
Using a label map.""" |
93
|
|
|
transform = RandomLabelsToImage( |
94
|
|
|
label_key='label', |
95
|
|
|
image_key='t1', |
96
|
|
|
used_labels=[1] |
97
|
|
|
) |
98
|
|
|
t1_indices = self.sample_subject['label'].data == 0 |
99
|
|
|
transformed = transform(self.sample_subject) |
100
|
|
|
self.assertTensorAlmostEqual( |
101
|
|
|
transformed['t1'].data[t1_indices], |
102
|
|
|
self.sample_subject['t1'].data[t1_indices] |
103
|
|
|
) |
104
|
|
|
|
105
|
|
|
def test_filling_with_discretized_label_map(self): |
106
|
|
|
"""The transform can fill in the generated image with an already |
107
|
|
|
existing image. |
108
|
|
|
Using a discretized label map.""" |
109
|
|
|
transform = RandomLabelsToImage( |
110
|
|
|
label_key='label', |
111
|
|
|
image_key='t1', |
112
|
|
|
discretize=True, |
113
|
|
|
used_labels=[1] |
114
|
|
|
) |
115
|
|
|
t1_indices = self.sample_subject['label'].data < 0.5 |
116
|
|
|
transformed = transform(self.sample_subject) |
117
|
|
|
self.assertTensorAlmostEqual( |
118
|
|
|
transformed['t1'].data[t1_indices], |
119
|
|
|
self.sample_subject['t1'].data[t1_indices] |
120
|
|
|
) |
121
|
|
|
|
122
|
|
|
def test_filling_with_discretized_pv_label_map(self): |
123
|
|
|
"""The transform can fill in the generated image with an already |
124
|
|
|
existing image. |
125
|
|
|
Using a discretized partial-volume label map.""" |
126
|
|
|
subject = self.get_subject_with_partial_volume_label_map(components=2) |
127
|
|
|
transform = RandomLabelsToImage( |
128
|
|
|
label_key='label', |
129
|
|
|
image_key='t1', |
130
|
|
|
discretize=True, |
131
|
|
|
used_labels=[1] |
132
|
|
|
) |
133
|
|
|
t1_indices = subject['label'].data.argmax(dim=0) == 0 |
134
|
|
|
transformed = transform(subject) |
135
|
|
|
self.assertTensorAlmostEqual( |
136
|
|
|
transformed['t1'].data[0][t1_indices], |
137
|
|
|
subject['t1'].data[0][t1_indices] |
138
|
|
|
) |
139
|
|
|
|
140
|
|
|
def test_filling_without_any_hole(self): |
141
|
|
|
"""The transform does not fill anything if there is no hole.""" |
142
|
|
|
transform = RandomLabelsToImage( |
143
|
|
|
label_key='label', |
144
|
|
|
image_key='t1', |
145
|
|
|
default_std=0, |
146
|
|
|
default_mean=-1, |
147
|
|
|
) |
148
|
|
|
original_t1 = self.sample_subject.t1.data.clone() |
149
|
|
|
transformed = transform(self.sample_subject) |
150
|
|
|
self.assertTensorNotEqual(original_t1, transformed.t1.data) |
151
|
|
|
|
152
|
|
|
def test_with_bad_default_mean_range(self): |
153
|
|
|
"""The transform raises an error if default_mean is not a |
154
|
|
|
single value nor a tuple of two values.""" |
155
|
|
|
with self.assertRaises(ValueError): |
156
|
|
|
RandomLabelsToImage(label_key='label', default_mean=(0, 1, 2)) |
157
|
|
|
|
158
|
|
|
def test_with_bad_default_mean_type(self): |
159
|
|
|
"""The transform raises an error if default_mean has the wrong type.""" |
160
|
|
|
with self.assertRaises(ValueError): |
161
|
|
|
RandomLabelsToImage(label_key='label', default_mean='wrong') |
162
|
|
|
|
163
|
|
|
def test_with_bad_default_std_range(self): |
164
|
|
|
"""The transform raises an error if default_std is not a |
165
|
|
|
single value nor a tuple of two values.""" |
166
|
|
|
with self.assertRaises(ValueError): |
167
|
|
|
RandomLabelsToImage(label_key='label', default_std=(0, 1, 2)) |
168
|
|
|
|
169
|
|
|
def test_with_bad_default_std_type(self): |
170
|
|
|
"""The transform raises an error if default_std has the wrong type.""" |
171
|
|
|
with self.assertRaises(ValueError): |
172
|
|
|
RandomLabelsToImage(label_key='label', default_std='wrong') |
173
|
|
|
|
174
|
|
|
def test_with_wrong_label_key_type(self): |
175
|
|
|
"""The transform raises an error if a wrong type is given for |
176
|
|
|
label_key.""" |
177
|
|
|
with self.assertRaises(TypeError): |
178
|
|
|
RandomLabelsToImage(label_key=42) |
179
|
|
|
|
180
|
|
|
def test_with_wrong_used_labels_type(self): |
181
|
|
|
"""The transform raises an error if a wrong type is given for |
182
|
|
|
used_labels.""" |
183
|
|
|
with self.assertRaises(TypeError): |
184
|
|
|
RandomLabelsToImage(label_key='label', used_labels=42) |
185
|
|
|
|
186
|
|
|
def test_with_wrong_used_labels_elements_type(self): |
187
|
|
|
"""The transform raises an error if wrong type are given for |
188
|
|
|
used_labels elements.""" |
189
|
|
|
with self.assertRaises(ValueError): |
190
|
|
|
RandomLabelsToImage(label_key='label', used_labels=['wrong']) |
191
|
|
|
|
192
|
|
|
def test_with_wrong_mean_type(self): |
193
|
|
|
"""The transform raises an error if wrong type is given for mean.""" |
194
|
|
|
with self.assertRaises(TypeError): |
195
|
|
|
RandomLabelsToImage(label_key='label', mean=42) |
196
|
|
|
|
197
|
|
|
def test_with_wrong_mean_elements_type(self): |
198
|
|
|
"""The transform raises an error if wrong type are given for |
199
|
|
|
mean elements.""" |
200
|
|
|
with self.assertRaises(ValueError): |
201
|
|
|
RandomLabelsToImage(label_key='label', mean=['wrong']) |
202
|
|
|
|
203
|
|
|
def test_with_wrong_std_type(self): |
204
|
|
|
"""The transform raises an error if wrong type is given for std.""" |
205
|
|
|
with self.assertRaises(TypeError): |
206
|
|
|
RandomLabelsToImage(label_key='label', std=42) |
207
|
|
|
|
208
|
|
|
def test_with_wrong_std_elements_type(self): |
209
|
|
|
"""The transform raises an error if wrong type are given for |
210
|
|
|
std elements.""" |
211
|
|
|
with self.assertRaises(ValueError): |
212
|
|
|
RandomLabelsToImage(label_key='label', std=['wrong']) |
213
|
|
|
|
214
|
|
|
def test_mean_and_std_len_not_matching(self): |
215
|
|
|
"""The transform raises an error if mean and std length don't match.""" |
216
|
|
|
with self.assertRaises(AssertionError): |
217
|
|
|
RandomLabelsToImage(label_key='label', mean=[0], std=[0, 1]) |
218
|
|
|
|
219
|
|
|
def test_mean_and_used_labels_len_not_matching(self): |
220
|
|
|
"""The transform raises an error if mean and used_labels |
221
|
|
|
length don't match.""" |
222
|
|
|
with self.assertRaises(AssertionError): |
223
|
|
|
RandomLabelsToImage( |
224
|
|
|
label_key='label', |
225
|
|
|
mean=[0], |
226
|
|
|
used_labels=[0, 1], |
227
|
|
|
) |
228
|
|
|
|
229
|
|
|
def test_std_and_used_labels_len_not_matching(self): |
230
|
|
|
"""The transform raises an error if std and used_labels |
231
|
|
|
length don't match.""" |
232
|
|
|
with self.assertRaises(AssertionError): |
233
|
|
|
RandomLabelsToImage(label_key='label', std=[0], used_labels=[0, 1]) |
234
|
|
|
|
235
|
|
|
def test_mean_not_matching_number_of_labels(self): |
236
|
|
|
"""The transform raises an error at runtime if mean length |
237
|
|
|
does not match label numbers.""" |
238
|
|
|
transform = RandomLabelsToImage(label_key='label', mean=[0]) |
239
|
|
|
with self.assertRaises(RuntimeError): |
240
|
|
|
transform(self.sample_subject) |
241
|
|
|
|
242
|
|
|
def test_std_not_matching_number_of_labels(self): |
243
|
|
|
"""The transform raises an error at runtime if std length |
244
|
|
|
does not match label numbers.""" |
245
|
|
|
transform = RandomLabelsToImage(label_key='label', std=[1, 2, 3]) |
246
|
|
|
with self.assertRaises(RuntimeError): |
247
|
|
|
transform(self.sample_subject) |
248
|
|
|
|
249
|
|
|
def test_bad_range(self): |
250
|
|
|
with self.assertRaises(ValueError): |
251
|
|
|
RandomLabelsToImage(default_mean=(2, 1)) |
252
|
|
|
|
253
|
|
|
def test_no_labels(self): |
254
|
|
|
transform = RandomLabelsToImage() |
255
|
|
|
with self.assertRaises(RuntimeError): |
256
|
|
|
transform(self.sample_subject.t1) |
257
|
|
|
|