|
1
|
|
|
from typing import Optional, Generator |
|
2
|
|
|
|
|
3
|
|
|
import numpy as np |
|
4
|
|
|
|
|
5
|
|
|
from ...typing import TypePatchSize, TypeTripletInt |
|
6
|
|
|
from ...data.subject import Subject |
|
7
|
|
|
from ...utils import to_tuple |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
class PatchSampler: |
|
11
|
|
|
r"""Base class for TorchIO samplers. |
|
12
|
|
|
|
|
13
|
|
|
Args: |
|
14
|
|
|
patch_size: Tuple of integers :math:`(w, h, d)` to generate patches |
|
15
|
|
|
of size :math:`w \times h \times d`. |
|
16
|
|
|
If a single number :math:`n` is provided, :math:`w = h = d = n`. |
|
17
|
|
|
|
|
18
|
|
|
.. warning:: This is an abstract class that should only be instantiated |
|
19
|
|
|
using child classes such as :class:`~torchio.data.UniformSampler` and |
|
20
|
|
|
:class:`~torchio.data.WeightedSampler`. |
|
21
|
|
|
""" |
|
22
|
|
|
def __init__(self, patch_size: TypePatchSize): |
|
23
|
|
|
patch_size_array = np.array(to_tuple(patch_size, length=3)) |
|
24
|
|
|
for n in patch_size_array: |
|
25
|
|
|
if n < 1 or not isinstance(n, (int, np.integer)): |
|
26
|
|
|
message = ( |
|
27
|
|
|
'Patch dimensions must be positive integers,' |
|
28
|
|
|
f' not {patch_size_array}' |
|
29
|
|
|
) |
|
30
|
|
|
raise ValueError(message) |
|
31
|
|
|
self.patch_size = patch_size_array.astype(np.uint16) |
|
32
|
|
|
|
|
33
|
|
|
def extract_patch( |
|
34
|
|
|
self, |
|
35
|
|
|
subject: Subject, |
|
36
|
|
|
index_ini: TypeTripletInt, |
|
37
|
|
|
) -> Subject: |
|
38
|
|
|
cropped_subject = self.crop(subject, index_ini, self.patch_size) |
|
39
|
|
|
cropped_subject['index_ini'] = np.array(index_ini).astype(int) |
|
40
|
|
|
return cropped_subject |
|
41
|
|
|
|
|
42
|
|
|
def crop( |
|
43
|
|
|
self, |
|
44
|
|
|
subject: Subject, |
|
45
|
|
|
index_ini: TypeTripletInt, |
|
46
|
|
|
patch_size: TypeTripletInt, |
|
47
|
|
|
) -> Subject: |
|
48
|
|
|
transform = self._get_crop_transform(subject, index_ini, patch_size) |
|
49
|
|
|
return transform(subject) |
|
50
|
|
|
|
|
51
|
|
|
@staticmethod |
|
52
|
|
|
def _get_crop_transform( |
|
53
|
|
|
subject, |
|
54
|
|
|
index_ini: TypeTripletInt, |
|
55
|
|
|
patch_size: TypePatchSize, |
|
56
|
|
|
): |
|
57
|
|
|
from ...transforms.preprocessing.spatial.crop import Crop |
|
58
|
|
|
shape = np.array(subject.spatial_shape, dtype=np.uint16) |
|
59
|
|
|
index_ini = np.array(index_ini, dtype=np.uint16) |
|
60
|
|
|
patch_size = np.array(patch_size, dtype=np.uint16) |
|
61
|
|
|
assert len(index_ini) == 3 |
|
62
|
|
|
assert len(patch_size) == 3 |
|
63
|
|
|
index_fin = index_ini + patch_size |
|
64
|
|
|
crop_ini = index_ini.tolist() |
|
65
|
|
|
crop_fin = (shape - index_fin).tolist() |
|
66
|
|
|
start = () |
|
67
|
|
|
cropping = sum(zip(crop_ini, crop_fin), start) |
|
68
|
|
|
return Crop(cropping) |
|
69
|
|
|
|
|
70
|
|
|
def __call__( |
|
71
|
|
|
self, |
|
72
|
|
|
subject: Subject, |
|
73
|
|
|
num_patches: Optional[int] = None, |
|
74
|
|
|
) -> Generator[Subject, None, None]: |
|
75
|
|
|
subject.check_consistent_space() |
|
76
|
|
|
if np.any(self.patch_size > subject.spatial_shape): |
|
77
|
|
|
message = ( |
|
78
|
|
|
f'Patch size {tuple(self.patch_size)} cannot be' |
|
79
|
|
|
f' larger than image size {tuple(subject.spatial_shape)}' |
|
80
|
|
|
) |
|
81
|
|
|
raise RuntimeError(message) |
|
82
|
|
|
kwargs = {} if num_patches is None else {'num_patches': num_patches} |
|
83
|
|
|
return self._generate_patches(subject, **kwargs) |
|
84
|
|
|
|
|
85
|
|
|
def _generate_patches( |
|
86
|
|
|
self, |
|
87
|
|
|
subject: Subject, |
|
88
|
|
|
num_patches: Optional[int] = None, |
|
89
|
|
|
) -> Generator[Subject, None, None]: |
|
90
|
|
|
raise NotImplementedError |
|
91
|
|
|
|
|
92
|
|
|
|
|
93
|
|
|
class RandomSampler(PatchSampler): |
|
94
|
|
|
r"""Base class for random samplers. |
|
95
|
|
|
|
|
96
|
|
|
Args: |
|
97
|
|
|
patch_size: Tuple of integers :math:`(w, h, d)` to generate patches |
|
98
|
|
|
of size :math:`w \times h \times d`. |
|
99
|
|
|
If a single number :math:`n` is provided, :math:`w = h = d = n`. |
|
100
|
|
|
""" |
|
101
|
|
|
def get_probability_map(self, subject: Subject): |
|
102
|
|
|
raise NotImplementedError |
|
103
|
|
|
|
fepegar /
torchio
