Newer
Older
# ============================================================================================= #
# Author: Filip Lux #
# Copyright: Filip Lux : lux.filip@gmail.com #
# #
# MIT License. #
# #
# Permission is hereby granted, free of charge, to any person obtaining a copy #
# of this software and associated documentation files (the "Software"), to deal #
# in the Software without restriction, including without limitation the rights #
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell #
# copies of the Software, and to permit persons to whom the Software is #
# furnished to do so, subject to the following conditions: #
# #
# The above copyright notice and this permission notice shall be included in all #
# copies or substantial portions of the Software. #
# #
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR #
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, #
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE #
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER #
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, #
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE #
# SOFTWARE. #
# ============================================================================================= #
from bio_volumentations.augmentations.transforms import (
GaussianNoise, PoissonNoise, Resize, Pad, Scale, Flip, CenterCrop, AffineTransform,
RandomScale, RandomRotate90, RandomFlip, RandomCrop, RandomAffineTransform, RandomGamma,
NormalizeMeanStd, GaussianBlur, Normalize, HistogramEqualization, RandomBrightnessContrast)
from bio_volumentations.core.composition import Compose
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
def get_keypoints_tests(transform,
in_shape: tuple = (32, 33, 34),
params: dict = {}):
w, h, d = in_shape
img = np.zeros((4, w, h, d), np.float32)
mask = np.zeros((w, h, d), np.int32)
keypoints = []
lbd = 3
for _ in range(15):
w1, h1, d1 = np.random.randint(lbd, w - lbd), \
np.random.randint(lbd, h - lbd), \
np.random.randint(lbd, d - lbd)
img[:, w1 - lbd:w1 + lbd, h1 - lbd:h1 + lbd, d1 - lbd:d1 + lbd] = 10.
mask[w1 - lbd:w1 + lbd, h1 - lbd:h1 + lbd, d1 - lbd:d1 + lbd] = 10.
keypoints.append((w1-0., h1-0., d1-0.))
sample = {'image': img,
'mask': mask,
'keypoints': keypoints}
tr = Compose([transform(**params, p=1)])
sample_transformed = tr(**sample)
keypoints_transformed = sample_transformed['keypoints']
if DEBUG:
print('KEYPOINTS', transform, keypoints)
print('KEYPOINTS TRANSFORMED', transform, keypoints_transformed)
tests = []
for k in keypoints_transformed:
coos = (np.array(k) + .5).astype(int)
tests.append((sample_transformed['image'][0, coos[0], coos[1], coos[2]], 10.,
f'mask, {k} {coos} {transform} {params}'))
tests.append((sample_transformed['mask'][coos[0], coos[1], coos[2]], 10.,
f'img {k} {coos} {transform}, {params}'))
return tests
def get_shape_tests(transform,
in_shape: tuple,
params={}):
"""
Iterates over all the possibilities, hot the array can passed throught the transform
Args:
transform: biovol transform,
in_shape: spatial dimension of the input image
params: optional, params of the biovol transform
Returns:
list of outputs and expected shapes
"""
w, h, d = in_shape
w_, h_, d_ = params['shape'] if 'shape' in params.keys() else (w, h, d)
res = []
tr = Compose([transform(**params, p=1)])
# img (W, H, D), mask (W, H, D)
img = np.ones((w, h, d), dtype=np.float32)
mask = np.ones((w, h, d), dtype=np.int32)
fmask = np.ones((w, h, d), dtype=np.float32)
#print(img.dtype, mask.dtype, fmask.dtype)
tr_img = tr(image=img, mask=mask, float_mask=fmask)
#print(tr_img['image'].dtype, tr_img['mask'].dtype, tr_img['float_mask'].dtype)
res.append((tr_img['image'], (1, w_, h_, d_), np.float32))
res.append((tr_img['mask'], (w_, h_, d_), np.int32))
res.append((tr_img['float_mask'], (w_, h_, d_), np.float32))
# img (C, W, H, D), mask (W, H, D)
img = np.ones((4, w, h, d), dtype=np.single)
mask = np.ones((w, h, d), dtype=int)
fmask = np.ones((w, h, d), dtype=np.single)
tr_img = tr(image=img, mask=mask, float_mask=fmask)
res.append((tr_img['image'], (4, w_, h_, d_), np.float32))
res.append((tr_img['mask'], (w_, h_, d_), np.int32))
res.append((tr_img['float_mask'], (w_, h_, d_), np.float32))
# img (C, W, H, D, T), mask (W, H, D, T)
img = np.ones((4, w, h, d, 5), dtype=np.single)
mask = np.ones((w, h, d, 5), dtype=int)
fmask = np.ones((w, h, d, 5), dtype=np.single)
tr_img = tr(image=img, mask=mask, float_mask=fmask)
res.append((tr_img['image'], (4, w_, h_, d_, 5), np.float32))
res.append((tr_img['mask'], (w_, h_, d_, 5), np.int32))
res.append((tr_img['float_mask'], (w_, h_, d_, 5), np.float32))
return res
class TestScale(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(Scale, (31, 32, 33), params={'scales': 1.5})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
tests = get_shape_tests(Scale, (31, 32, 33), params={'scales': 0.8})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
tests = get_keypoints_tests(Scale, params={'scales': 1.5})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
tests = get_keypoints_tests(Scale, params={'scales': 0.8})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
class TestRandomScale(unittest.TestCase):
def test_shape(self):
limits = [0.2,
(0.8, 1.2),
(0.2, 0.3, 0.1),
(0.8, 1.2, 0.9, 1.1, 0.7, 1.)]
tests = get_shape_tests(RandomScale,
in_shape=(31, 32, 33),
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
limits = [0.2,
(0.8, 1.2),
(0.2, 0.3, 0.1),
(0.8, 1.2, 0.9, 1.1, 0.7, 1.)]
for scaling_limit in limits:
tests = get_keypoints_tests(RandomScale,
in_shape=(61, 62, 63),
params={'scaling_limit': scaling_limit})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
class TestRandomRotate90(unittest.TestCase):
def test_shape(self):
axes_list = [None,
[1],
[1, 2],
[1, 2, 3]]
for axes in axes_list:
tests = get_shape_tests(RandomRotate90, (30, 30, 30),
params={'axes': axes})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
axes_list = [None,
[1],
[1, 2],
[1, 2, 3]]
for axes in axes_list:
tests = get_keypoints_tests(RandomRotate90, params={'axes': axes})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
class TestFlip(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(Flip, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
tests = get_keypoints_tests(Flip, params={})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
class TestRandomFlip(unittest.TestCase):
def test_shape(self):
axes_list = [None,
[],
[1],
[1, 2],
[1, 2, 3]]
for axes in axes_list:
tests = get_shape_tests(RandomFlip, (30, 30, 30),
params={'axes_to_choose': axes})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
axes_list = [None,
[],
[1],
[1, 2],
[1, 2, 3]]
for axes in axes_list:
tests = get_keypoints_tests(RandomFlip, params={'axes_to_choose': axes})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
class TestCenterCrop(unittest.TestCase):
def test_inflate(self):
in_shape = (32, 31, 30)
shape_tests = get_shape_tests(CenterCrop, in_shape, {'shape': (40, 41, 43)})
shape_tests = get_shape_tests(CenterCrop, in_shape, {'shape': (20, 21, 23)})
def test_keypoints(self):
in_shape = (32, 31, 30)
tests = get_keypoints_tests(CenterCrop, in_shape, params={'shape': (40, 41, 42)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(CenterCrop, in_shape, params={'shape': (20, 21, 22)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
class TestRandomCrop(unittest.TestCase):
def test_inflate(self):
in_shape = (32, 31, 30)
shape_tests = get_shape_tests(RandomCrop, in_shape, {'shape': (40, 41, 42)})
def test_deflate(self):
in_shape = (32, 31, 30)
shape_tests = get_shape_tests(RandomCrop, in_shape, {'shape': (20, 21, 22)})
def test_keypoints(self):
in_shape = (32, 31, 30)
tests = get_keypoints_tests(RandomCrop, in_shape, params={'shape': (40, 41, 42)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(RandomCrop, in_shape, params={'shape': (20, 21, 22)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
class TestResize(unittest.TestCase):
def test_inflate(self):
in_shape = (32, 31, 30)
shape_tests = get_shape_tests(Resize, in_shape, {'shape': (40, 41, 42)})
def test_deflate(self):
in_shape = (32, 31, 30)
shape_tests = get_shape_tests(Resize, in_shape, {'shape': (20, 21, 22)})
def test_keypoints(self):
in_shape = (32, 31, 30)
tests = get_keypoints_tests(Resize, in_shape, params={'shape': (40, 41, 42)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(Resize, in_shape, params={'shape': (20, 21, 22)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
class TestPad(unittest.TestCase):
def test_1(self):
tr = Compose([Pad(2)])
img = np.empty((30, 30, 30))
tr_img = tr(image=img)['image']
self.assertTupleEqual(tr_img.shape, (1, 34, 34, 34))
img = np.empty((1, 30, 30, 30))
tr_img = tr(image=img)['image']
self.assertTupleEqual(tr_img.shape, (1, 34, 34, 34))
img = np.empty((4, 30, 30, 30))
tr_img = tr(image=img)['image']
self.assertTupleEqual(tr_img.shape, (4, 34, 34, 34))
img = np.empty((4, 30, 30, 30, 5))
tr_img = tr(image=img)['image']
self.assertTupleEqual(tr_img.shape, (4, 34, 34, 34, 5))
def test_keypoints(self):
in_shape = (32, 31, 30)
tests = get_keypoints_tests(Pad, in_shape, params={'pad_size': (5, 8)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(Pad, in_shape, params={'pad_size': 4})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(Pad, in_shape, params={'pad_size': (3, 4, 5, 6, 7, 8)})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
class TestRandomAffineTransform(unittest.TestCase):
def test_shape(self):
angle_limits = [10,
(-20, 20),
(12, 30, 0),
(-20, 20, -180, 180, 0, 0)]
for angle_limit in angle_limits:
tests = get_shape_tests(RandomAffineTransform, (31, 32, 33),
params={'angle_limit': angle_limit})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
scale_limits = [0.2,
(0.8, 1.2),
(0.2, 0.3, 0.1),
(0.8, 1.2, 0.9, 1.1, 0.7, 1.)]
for scale_limit in scale_limits:
tests = get_shape_tests(RandomAffineTransform, (31, 32, 33),
params={'scaling_limit': scale_limit})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
translation_limits = [10,
(0, 12),
(3, 5, 10),
(-3, 3, -5, 5, 0, 0)]
for translation in translation_limits:
tests = get_shape_tests(RandomAffineTransform, (31, 32, 33),
params={'translation_limit': translation})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
def test_keypoints(self):
in_shape = (61, 62, 63)
angle_limits = [10,
(-20, 20),
(12, 30, 0),
(-20, 20, -180, 180, 0, 0)]
for angle_limit in angle_limits:
tests = get_keypoints_tests(RandomAffineTransform,
in_shape=in_shape,
params={'angle_limit': angle_limit})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.1, msg)
scale_limits = [0.2,
(0.8, 1.2),
(0.2, 0.3, 0.1),
(0.8, 1.2, 0.9, 1.1, 0.7, 1.)]
for scale_limit in scale_limits:
tests = get_keypoints_tests(RandomAffineTransform,
in_shape=in_shape,
params={'scaling_limit': scale_limit})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
translation_limits = [10,
(0, 12),
(3, 5, 10),
(-3, 3, -5, 5, 0, 0)]
for translation in translation_limits:
tests = get_keypoints_tests(RandomAffineTransform,
in_shape=in_shape,
params={'translation_limit': translation})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.2, msg)
class TestAffineTransform(unittest.TestCase):
def test_shape(self):
scale = (1.2, 0.8, 1)
translation = (0, 1, -40)
angles = (-20, 0, -0.5)
tests = get_shape_tests(AffineTransform, (31, 32, 33),
params={'translation': translation})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
tests = get_shape_tests(AffineTransform, (31, 32, 33),
params={'scale': scale})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
tests = get_shape_tests(AffineTransform, (31, 32, 33),
params={'angles': angles})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
def test_keypoints(self):
scale = (1.2, 0.8, 1)
translation = (0, 1, -40)
angles = (-20, 0, -0.5)
tests = get_keypoints_tests(AffineTransform,
in_shape=(61, 62, 63),
params={'scale': scale})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(AffineTransform,
in_shape=(61, 62, 63),
params={'translation': translation})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
tests = get_keypoints_tests(AffineTransform,
in_shape=(61, 62, 63),
params={'angles': angles})
for value, expected_value, msg in tests:
self.assertGreater(value, expected_value * 0.5, msg)
# ImageTransforms
class TestNormalizeMeanStd(unittest.TestCase):
def test_shape(self):
mean = 1.2
std = 2
tests = get_shape_tests(NormalizeMeanStd, (31, 32, 33),
params={'mean': mean,
'std': std})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestGaussianNoise(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(GaussianNoise, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestPoissonNoise(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(PoissonNoise, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
class TestGaussianBlur(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(GaussianBlur, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestRandomGamma(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(RandomGamma, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestRandomBrightnessContrast(unittest.TestCase):
def test_shape(self):
brightness_list = [3, (2, 5)]
contrast_list = [0.5, (.7, 1.1)]
for brightness in brightness_list:
for contrast in contrast_list:
tests = get_shape_tests(RandomBrightnessContrast, (30, 31, 32),
params={'brightness_limit': brightness,
'contrast_limit': contrast})
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestHistogramEqualization(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(HistogramEqualization, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)
class TestNormalize(unittest.TestCase):
def test_shape(self):
tests = get_shape_tests(Normalize, (31, 32, 33))
for tr_img, expected_shape, data_type in tests:
self.assertTupleEqual(tr_img.shape, expected_shape)
self.assertEqual(tr_img.dtype, data_type)