|
1
|
|
|
from typing import Tuple, Optional, List |
|
2
|
|
|
import torch |
|
3
|
|
|
from ....torchio import DATA, TypeRangeFloat |
|
4
|
|
|
from ....data.subject import Subject |
|
5
|
|
|
from .. import RandomTransform |
|
6
|
|
|
|
|
7
|
|
|
|
|
8
|
|
|
class RandomGamma(RandomTransform): |
|
9
|
|
|
r"""Change contrast of an image by raising its values to the power |
|
10
|
|
|
:math:`\gamma`. |
|
11
|
|
|
|
|
12
|
|
|
Args: |
|
13
|
|
|
log_gamma: Tuple :math:`(a, b)` to compute the exponent |
|
14
|
|
|
:math:`\gamma = e ^ \beta`, |
|
15
|
|
|
where :math:`\beta \sim \mathcal{U}(a, b)`. |
|
16
|
|
|
If a single value :math:`d` is provided, then |
|
17
|
|
|
:math:`\beta \sim \mathcal{U}(-d, d)`. |
|
18
|
|
|
Negative and positive values for this argument perform gamma |
|
19
|
|
|
compression and expansion, respectively. |
|
20
|
|
|
See the `Gamma correction`_ Wikipedia entry for more information. |
|
21
|
|
|
p: Probability that this transform will be applied. |
|
22
|
|
|
seed: See :py:class:`~torchio.transforms.augmentation.RandomTransform`. |
|
23
|
|
|
keys: See :py:class:`~torchio.transforms.Transform`. |
|
24
|
|
|
|
|
25
|
|
|
.. _Gamma correction: https://en.wikipedia.org/wiki/Gamma_correction |
|
26
|
|
|
|
|
27
|
|
|
Example: |
|
28
|
|
|
>>> import torchio |
|
29
|
|
|
>>> from torchio import RandomGamma |
|
30
|
|
|
>>> from torchio.datasets import FPG |
|
31
|
|
|
>>> sample = FPG() |
|
32
|
|
|
>>> transform = RandomGamma(log_gamma=(-0.3, 0.3)) # gamma between 0.74 and 1.34 |
|
33
|
|
|
>>> transformed = transform(sample) |
|
34
|
|
|
""" |
|
35
|
|
|
def __init__( |
|
36
|
|
|
self, |
|
37
|
|
|
log_gamma: TypeRangeFloat = (-0.3, 0.3), |
|
38
|
|
|
p: float = 1, |
|
39
|
|
|
seed: Optional[int] = None, |
|
40
|
|
|
keys: Optional[List[str]] = None, |
|
41
|
|
|
): |
|
42
|
|
|
super().__init__(p=p, seed=seed, keys=keys) |
|
43
|
|
|
self.log_gamma_range = self.parse_range(log_gamma, 'log_gamma') |
|
44
|
|
|
|
|
45
|
|
|
def apply_transform(self, sample: Subject) -> dict: |
|
46
|
|
|
random_parameters_images_dict = {} |
|
47
|
|
|
for image_name, image_dict in sample.get_images_dict().items(): |
|
48
|
|
|
gamma = self.get_params(self.log_gamma_range) |
|
49
|
|
|
random_parameters_dict = {'gamma': gamma} |
|
50
|
|
|
random_parameters_images_dict[image_name] = random_parameters_dict |
|
51
|
|
|
image_dict[DATA] = image_dict[DATA] ** gamma |
|
|
|
|
|
|
52
|
|
|
sample.add_transform(self, random_parameters_images_dict) |
|
53
|
|
|
return sample |
|
54
|
|
|
|
|
55
|
|
|
@staticmethod |
|
56
|
|
|
def get_params(log_gamma_range: Tuple[float, float]) -> torch.Tensor: |
|
57
|
|
|
gamma = torch.FloatTensor(1).uniform_(*log_gamma_range).exp() |
|
58
|
|
|
return gamma |
|
59
|
|
|
|
fepegar /
torchio
Loading history...