Newer
Older
# ============================================================================================= #
# Author: Pavel Iakubovskii, ZFTurbo, ashawkey, Dominik Müller, #
# Samuel Šuľan, Lucia Hradecká, Filip Lux #
# Copyright: albumentations: : https://github.com/albumentations-team #
# Pavel Iakubovskii : https://github.com/qubvel #
# ZFTurbo : https://github.com/ZFTurbo #
# ashawkey : https://github.com/ashawkey #
# Dominik Müller : https://github.com/muellerdo #
# Lucia Hradecká : lucia.d.hradecka@gmail.com #
# Filip Lux : lux.filip@gmail.com #
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
# #
# Volumentations History: #
# - Original: https://github.com/albumentations-team/albumentations #
# - 3D Conversion: https://github.com/ashawkey/volumentations #
# - Continued Development: https://github.com/ZFTurbo/volumentations #
# - Enhancements: https://github.com/qubvel/volumentations #
# - Further Enhancements: https://github.com/muellerdo/volumentations #
# - Biomedical Enhancements: https://gitlab.fi.muni.cz/cbia/bio-volumentations #
# #
# 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. #
# ============================================================================================= #
import random
import numpy as np
from ..core.transforms_interface import DualTransform, ImageOnlyTransform
from ..augmentations import functional as F
from ..augmentations.spatial_funcional import get_affine_transform, parse_itk_interpolation
from ..random_utils import uniform, sample_range_uniform
from typing import List, Sequence, Tuple, Union
from .utils import parse_limits, parse_coefs, parse_pads, to_tuple, validate_bbox, get_spatio_temporal_domain_limit,\
to_spatio_temporal
# TODO anti_aliasing_downsample keep parameter or remove?
class Resize(DualTransform):
"""Resize input to the given shape.
Internally, the ``skimage.transform.resize`` function is used.
The ``interpolation``, ``border_mode``, ``ival``, ``mval``,
and ``anti_aliasing_downsample`` arguments are forwarded to it. More details at:
https://scikit-image.org/docs/stable/api/skimage.transform.html#skimage.transform.resize.
Args:
shape (tuple of ints): The desired image shape.
The unspecified dimensions (C and T) are not affected.
interpolation (int, optional): Order of spline interpolation.
Defaults to ``1``.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'reflect'``.
ival (float, optional): Value of `image` voxels outside of the `image` domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
mval (float, optional): Value of `mask` and `float_mask` voxels outside of the domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
anti_aliasing_downsample (bool, optional): Controls if the Gaussian filter should be applied before
downsampling. Recommended.
Defaults to ``True``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``1``.
Targets:
image, mask, float mask, key points, bounding boxes
def __init__(self, shape: TypeSpatialShape, interpolation: int = 1, border_mode: str = 'reflect', ival: float = 0,
mval: float = 0, anti_aliasing_downsample: bool = True, ignore_index: Union[float, None] = None,
always_apply: bool = False, p: float = 1):
super().__init__(always_apply, p)
self.shape: TypeSpatioTemporalCoordinate = to_spatio_temporal(shape)
self.interpolation = interpolation
self.border_mode = border_mode
self.mask_mode = border_mode
self.ival = ival
self.mval = mval
self.anti_aliasing_downsample = anti_aliasing_downsample
if not (ignore_index is None):
self.mask_mode = "constant"
self.mval = ignore_index
def apply(self, img, **params):
return F.resize(img, input_new_shape=self.shape, interpolation=self.interpolation,
border_mode=self.border_mode, cval=self.ival,
anti_aliasing_downsample=self.anti_aliasing_downsample)
def apply_to_mask(self, mask, **params):
return F.resize(mask, input_new_shape=self.shape, interpolation=0,
border_mode=self.mask_mode, cval=self.mval, anti_aliasing_downsample=False,
mask=True)
def apply_to_float_mask(self, mask, **params):
return F.resize(mask, input_new_shape=self.shape, interpolation=self.interpolation,
border_mode=self.mask_mode, cval=self.mval, anti_aliasing_downsample=False,
mask=True)
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
def apply_to_keypoints(self, keypoints, **params):
return F.resize_keypoints(keypoints,
domain_limit=params['domain_limit'],
new_shape=self.shape)
"""
def apply_to_bboxes(self, bboxes, **params):
for bbox in bboxes:
new_bbox = F.resize_keypoints(bbox,
input_new_shape=self.shape,
original_shape=params['original_shape'],
keep_all=True)
if validate_bbox(bbox, new_bbox, min_overlay_ratio):
res.append(new_bbox)
return res
"""
def get_params(self, **data):
# read shape of the original image
domain_limit: TypeSpatioTemporalCoordinate = get_spatio_temporal_domain_limit(data)
return {
"domain_limit": domain_limit,
}
def __repr__(self):
return f'Resize({self.shape}, {self.interpolation}, {self.border_mode} , {self.ival}, {self.mval},' \
f'{self.anti_aliasing_downsample}, {self.always_apply}, {self.p})'
"""Rescale the input image content by the given scale. The image shape remains unchanged.
Args:
scales (float|List[float], optional): Value by which the input should be scaled.
Must be either of: ``S``, ``[S_Z, S_Y, S_X]``.
If a float, then all spatial dimensions are scaled by it (equivalent to ``[S, S, S]``).
The unspecified dimensions (C and T) are not affected.
interpolation (str, optional): SimpleITK interpolation type for `image` and `float_mask`.
Must be one of ``linear``, ``nearest``, ``bspline``, ``gaussian``.
For `mask`, the ``nearest`` interpolation is always used.
spacing (float | Tuple[float, float, float] | None, optional): Voxel spacing for individual spatial dimensions.
Must be either of: ``S``, ``(S1, S2, S3)``, or ``None``.
If ``None``, equivalent to ``(1, 1, 1)``.
If a float ``S``, equivalent to ``(S, S, S)``.
Otherwise, a scale for each spatial dimension must be given.
Defaults to ``None``.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'constant'``.
ival (float, optional): Value of `image` voxels outside of the `image` domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
mval (float, optional): Value of `mask` and `float_mask` voxels outside of the domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``1``.
Targets:
image, mask, float mask, key points, bounding boxes
"""
def __init__(self, scales: Union[float, TypeTripletFloat] = 1,
interpolation: str = 'linear', spacing: Union[float, TypeTripletFloat] = None,
border_mode: str = 'constant', ival: float = 0, mval: float = 0,
ignore_index: Union[float, None] = None, always_apply: bool = False, p: float = 1):
super().__init__(always_apply, p)
self.scale = parse_coefs(scales, identity_element=1.)
self.interpolation: str = parse_itk_interpolation(interpolation)
self.spacing: TypeTripletFloat = parse_coefs(spacing, identity_element=1.)
self.border_mode = border_mode # not implemented
self.mask_mode = border_mode # not implemented
self.ival = ival
self.mval = mval
if not (ignore_index is None):
self.mask_mode = "constant"
self.mval = ignore_index
def apply(self, img, **params):
return F.affine(img,
scales=self.scale,
interpolation=self.interpolation,
border_mode=self.border_mode,
value=self.ival,
spacing=self.spacing)
def apply_to_mask(self, mask, **params):
interpolation = parse_itk_interpolation('nearest') # refers to 'sitkNearestNeighbor'
scales=self.scale,
interpolation=interpolation,
border_mode=self.mask_mode,
value=self.mval,
scales=self.scale,
interpolation=self.interpolation,
border_mode=self.mask_mode,
value=self.mval,
def apply_to_keypoints(self, keypoints, **params):
return F.affine_keypoints(keypoints,
scales=self.scale,
spacing = self.spacing,
domain_limit=params['domain_limit'])
"""
def apply_to_bboxes(self, bboxes, **params):
for bbox in bboxes:
new_bbox = F.affine_keypoints(bbox,
scales=self.scale,
domain_limit=params['domain_limit'],
spacing = self.spacing,
keep_all=True)
if validate_bbox(bbox, new_bbox):
res.append(new_bbox)
return res
"""
def get_params(self, **data):
domain_limit: TypeSpatioTemporalCoordinate = get_spatio_temporal_domain_limit(data)
return {'domain_limit': domain_limit}
def __repr__(self):
return f'Scale({self.scale}, {self.interpolation}, {self.border_mode}, {self.ival}, {self.mval},' \
f'{self.always_apply}, {self.p})'
class RandomScale(DualTransform):
"""Randomly rescale the input image content by the given scale. The image shape remains unchanged.
scaling_limit (float | Tuple[float], optional): Limits of scaling factors.
Must be either of: ``S``, ``(S1, S2)``, ``(S_Z, S_Y, S_X)``, or ``(S_Z1, S_Z2, S_Y1, S_Y2, S_X1, S_X2)``.
If a float ``S``, then all spatial dimensions are scaled by a random number drawn uniformly from
the interval [1/S, S] (equivalent to inputting ``(1/S, S, 1/S, S, 1/S, S)``).
If a tuple of 2 floats, then all spatial dimensions are scaled by a random number drawn uniformly
from the interval [S1, S2] (equivalent to inputting ``(S1, S2, S1, S2, S1, S2)``).
If a tuple of 3 floats, then an interval [1/S_a, S_a] is constructed for each spatial
(equivalent to inputting ``(1/S_Z, S_Z, 1/S_Y, S_Y, 1/S_X, S_X)``).
If a tuple of 6 floats, the scales for individual spatial dimensions are randomly drawn from the
respective intervals [S_Z1, S_Z2], [S_Y1, S_Y2], [S_X1, S_X2].
The unspecified dimensions (C and T) are not affected.
Defaults to ``(1.1)``.
interpolation (str, optional): SimpleITK interpolation type for `image` and `float_mask`.
Must be one of ``linear``, ``nearest``, ``bspline``, ``gaussian``.
For `mask`, the ``nearest`` interpolation is always used.
spacing (float | Tuple[float, float, float] | None, optional): Voxel spacing for individual spatial dimensions.
Must be either of: ``S``, ``(S1, S2, S3)``, or ``None``.
If ``None``, equivalent to ``(1, 1, 1)``.
If a float ``S``, equivalent to ``(S, S, S)``.
Otherwise, a scale for each spatial dimension must be given.
Defaults to ``None``.
border_mode (str, optional): Values outside image domain are filled according to the mode.
Defaults to ``'constant'``.
ival (float, optional): Value of `image` voxels outside of the `image` domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
mval (float, optional): Value of `mask` and `float_mask` voxels outside of the domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``0.5``.
Targets:
image, mask, float mask, key points, bounding boxes
"""
def __init__(self, scaling_limit: Union[float, TypePairFloat, TypeTripletFloat, TypeSextetFloat] = (0.9, 1.1),
interpolation: str = 'linear', spacing: Union[float, TypeTripletFloat] = None,
border_mode: str = 'constant', ival: float = 0, mval: float = 0,
ignore_index: Union[float, None] = None, always_apply: bool = False, p: float = 0.5):
super().__init__(always_apply, p)
self.scaling_limit: TypeSextetFloat = parse_limits(scaling_limit, scale=True)
self.interpolation: str = parse_itk_interpolation(interpolation)
self.spacing: TypeTripletFloat = parse_coefs(spacing, identity_element=1.)
self.border_mode = border_mode
self.mask_mode = border_mode
self.ival: float = ival
self.mval: float = mval
if not (ignore_index is None):
self.mask_mode = "constant"
self.mval = ignore_index
def get_params(self, **data):
# set parameters of the transform
domain_limit: TypeSpatioTemporalCoordinate = get_spatio_temporal_domain_limit(data)
scale = sample_range_uniform(self.scaling_limit)
return {
"scale": scale,
}
def apply(self, img, **params):
return F.affine(img,
scales=params["scale"],
interpolation=self.interpolation,
border_mode=self.border_mode,
value=self.ival,
spacing=self.spacing)
def apply_to_mask(self, mask, **params):
interpolation = parse_itk_interpolation('nearest') # refers to 'sitkNearestNeighbor'
scales=params["scale"],
interpolation=interpolation,
border_mode=self.mask_mode,
value=self.mval,
scales=params["scale"],
interpolation=self.interpolation,
border_mode=self.mask_mode,
value=self.mval,
def apply_to_keypoints(self, keypoints, **params):
return F.affine_keypoints(keypoints,
scales=params["scale"],
spacing=self.spacing,
domain_limit=params['domain_limit'])
def __repr__(self):
return f'RandomScale({self.scaling_limit}, {self.interpolation}, {self.always_apply}, {self.p})'
class RandomRotate90(DualTransform):
"""Rotation of input by 0, 90, 180, or 270 degrees around the specified spatial axes.
Args:
axes (List[int], optional): List of axes around which the input is rotated. Recognised axis symbols are
``1`` for Z, ``2`` for Y, and ``3`` for X. A single axis can occur multiple times in the list.
If ``shuffle_axis = False``, the order of axes determines the order of transformations.
If ``None``, will be rotated around all spatial axes.
shuffle_axis (bool, optional): If set to ``True``, the order of rotations is random.
Defaults to ``False``.
factor (int, optional): Number of times the array is rotated by 90 degrees. If ``None``, will be chosen randomly.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``0.5``.
Targets:
image, mask, float mask, key points, bounding boxes
def __init__(self, axes: List[int] = None, shuffle_axis: bool = False, factor=None,
always_apply: bool = False, p: float = 0.5):
super().__init__(always_apply, p)
self.axes = axes
self.shuffle_axis = shuffle_axis
def apply(self, img, **params):
for factor, axes in zip(params["factor"], params["rotation_around"]):
img = np.rot90(img, factor, axes=axes)
return img
def apply_to_mask(self, mask, **params):
for rot, factor in zip(params["rotation_around"], params["factor"]):
mask = np.rot90(mask, factor, axes=(rot[0] - 1, rot[1] - 1))
def apply_to_keypoints(self, keypoints, **params):
for rot, factor in zip(params["rotation_around"], params["factor"]):
keypoints = F.rot90_keypoints(keypoints,
factor=factor,
def get_params(self, **data):
# Rotate by all axis by default
if self.axes is None:
self.axes = [1, 2, 3]
# Create all combinations for rotating
axes_to_rotate = {1: (2, 3), 2: (1, 3), 3: (1, 2)}
rotation_around = []
for i in self.axes:
if i in axes_to_rotate.keys():
rotation_around.append(axes_to_rotate[i])
# shuffle order of axis
if self.shuffle_axis:
random.shuffle(rotation_around)
# choose angle to rotate
if self.factor is None:
factor = [random.randint(0, 3) for _ in range(len(rotation_around))]
else:
factor = [self.factor]
rotation_around = [(1, 2)]
print('ROT90', factor, rotation_around)
"rotation_around": rotation_around,
"img_shape": img_shape}
def __repr__(self):
return f'RandomRotate90({self.axes}, {self.always_apply}, {self.p})'
class Flip(DualTransform):
"""Flip input around the specified spatial axes.
Args:
axes (List[int], optional): List of axes around which is flip done. Recognised axis symbols are
``1`` for Z, ``2`` for Y, and ``3`` for X. If ``None``, will be flipped around all spatial axes.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``1``.
Targets:
image, mask, float mask, key points, bounding boxes
"""
def __init__(self, axes: List[int] = None, always_apply=False, p=1):
super().__init__(always_apply, p)
self.axes = axes
def apply(self, img, **params):
return np.flip(img, params["axes"])
def apply_to_mask(self, mask, **params):
# Mask has no dimension channel
return np.flip(mask, axis=[item - 1 for item in params["axes"]])
def apply_to_keypoints(self, keypoints, **params):
return F.flip_keypoints(keypoints,
axes=params['axes'],
img_shape=params['img_shape'])
axes = [1, 2, 3] if self.axes is None else self.axes
img_shape = np.array(data['image'].shape[1:4])
return {"axes": axes,
"img_shape": img_shape}
def __repr__(self):
return f'Flip({self.axes}, {self.always_apply}, {self.p})'
class RandomFlip(DualTransform):
"""Flip input around a set of axes randomly chosen from the input list of axis combinations.
Args:
axes_to_choose (List[Tuple[int]] or None, optional): List of axis indices from which one option
is randomly chosen. Recognised axis symbols are ``1`` for Z, ``2`` for Y, and ``3`` for X.
The image will be flipped around all axes in the chosen combination.
If ``None``, a random subset of spatial axes is chosen, corresponding to inputting
``[(,), (1,), (2,), (3,), (1, 2), (1, 3), (2, 3), (1, 2, 3)]``.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``0.5``.
Targets:
image, mask, float mask, key points, bounding boxes
"""
def __init__(self, axes_to_choose: Union[None, List[Tuple[int]]] = None, always_apply=False, p=0.5):
super().__init__(always_apply, p)
# TODO: check if input value `axes_to_choose` valid
self.axes = axes_to_choose
def apply(self, img, **params):
return np.flip(img, params["axes"])
def apply_to_mask(self, mask, **params):
# Mask has no dimension channel
return np.flip(mask, axis=[item - 1 for item in params["axes"]])
def apply_to_keypoints(self, keypoints, keep_all=False, **params):
return F.flip_keypoints(keypoints,
axes=params['axes'],
img_shape=params['img_shape'])
to_choose = [1, 2, 3] if self.axes is None else self.axes
axes = random.sample(to_choose, random.randint(0, len(to_choose)))
img_shape = np.array(data['image'].shape[1:4])
return {"axes": axes,
"img_shape": img_shape}
return f'RandomFlip({self.axes}, {self.always_apply}, {self.p})'
class CenterCrop(DualTransform):
"""Crops the central region of the input of given size.
Unlike ``CenterCrop`` from `Albumentations`, this transform pads the input in dimensions
where the input is smaller than the ``shape`` with ``numpy.pad``. The ``border_mode``, ``ival`` and ``mval``
arguments are forwarded to ``numpy.pad`` if padding is necessary. More details at:
https://numpy.org/doc/stable/reference/generated/numpy.pad.html.
Args:
shape (Tuple[int]): The desired shape of input.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'reflect'``.
ival (float | Sequence, optional): Values of `image` voxels outside of the `image` domain.
Only applied when ``border_mode = 'constant'`` or ``border_mode = 'linear_ramp'``.
Defaults to ``(0, 0)``.
mval (float | Sequence, optional): Values of `mask` voxels outside of the `mask` domain.
Only applied when ``border_mode = 'constant'`` or ``border_mode = 'linear_ramp'``.
Defaults to ``(0, 0)``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``1``.
Targets:
image, mask, float mask, key points, bounding boxes
def __init__(self, shape: TypeSpatialShape, border_mode: str = "reflect", ival: Union[Sequence[float], float] = (0, 0),
mval: Union[Sequence[float], float] = (0, 0), ignore_index: Union[float, None] = None,
always_apply: bool = False, p: float = 1.0):
super().__init__(always_apply, p)
self.output_shape = np.asarray(shape, dtype=np.intc) # TODO: make it len 3
self.border_mode = border_mode
self.mask_mode = border_mode
self.ival = ival
self.mval = mval
if not (ignore_index is None):
self.mask_mode = "constant"
self.mval = ignore_index
def apply(self, img, **params):
return F.crop(img,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
border_mode=self.mask_mode, cval=self.mval, mask=False)
return F.crop(mask,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
border_mode=self.mask_mode, cval=self.mval, mask=True)
def apply_to_keypoints(self, keypoints, keep_all=False, **params):
return F.crop_keypoints(keypoints,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
keep_all=keep_all)
def get_params(self, **data):
# get crop coordinates, position of the corner closest to the image origin
img_spatial_shape = np.array(data['image'].shape[1:4])
position: TypeSpatialCoordinate = (img_spatial_shape - self.output_shape) // 2
position = np.maximum(position, 0).astype(int)
pad_dims = F.get_pad_dims(img_spatial_shape, self.output_shape)
return {'crop_position': position,
'pad_dims': pad_dims}
return f'CenterCrop({self.output_shape}, {self.always_apply}, {self.p})'
class RandomCrop(DualTransform):
"""Randomly crops a region of given size from the input.
Unlike ``RandomCrop`` from `Albumentations`, this transform pads the input in dimensions
where the input is smaller than the ``shape`` with ``numpy.pad``. The ``border_mode``, ``ival`` and ``mval``
arguments are forwarded to ``numpy.pad`` if padding is necessary. More details at:
https://numpy.org/doc/stable/reference/generated/numpy.pad.html.
Args:
shape (Tuple[int]): The desired shape of input.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'reflect'``.
ival (float | Sequence, optional): Values of `image` voxels outside of the `image` domain.
Only applied when ``border_mode = 'constant'`` or ``border_mode = 'linear_ramp'``.
Defaults to ``(0, 0)``.
mval (float | Sequence, optional): Values of `mask` voxels outside of the `mask` domain.
Only applied when ``border_mode = 'constant'`` or ``border_mode = 'linear_ramp'``.
Defaults to ``(0, 0)``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``1``.
Targets:
image, mask, float mask, key points, bounding boxes
def __init__(self, shape: TypeSpatialShape, border_mode: str = "reflect", ival: Union[Sequence[float], float] = (0, 0),
mval: Union[Sequence[float], float] = (0, 0), ignore_index: Union[float, None] = None,
always_apply: bool = False, p: float = 1.0):
super().__init__(always_apply, p)
self.border_mode = border_mode
self.mask_mode = border_mode
self.ival = ival
self.mval = mval
if not (ignore_index is None):
self.mask_mode = "constant"
self.mval = ignore_index
def apply(self, img, **params):
return F.crop(img,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
border_mode=self.mask_mode, cval=self.mval, mask=False)
def apply_to_mask(self, mask, **params):
return F.crop(mask,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
border_mode=self.mask_mode, cval=self.mval, mask=True)
def apply_to_keypoints(self, keypoints, keep_all=False, **params):
return F.crop_keypoints(keypoints,
crop_shape=self.output_shape,
crop_position=params['crop_position'],
pad_dims=params['pad_dims'],
keep_all=keep_all)
# get crop coordinates, position of the corner closest to the image origin
img_spatial_shape = np.array(data['image'].shape[1:4])
ranges: TypeSpatialShape = np.maximum(img_spatial_shape - self.output_shape, 0)
position = np.array([random.randint(0, r) for r in ranges])
pad_dims = F.get_pad_dims(img_spatial_shape, self.output_shape)
return {'crop_position': position,
'pad_dims': pad_dims}
return f'RandomCrop({self.output_shape}, {self.always_apply}, {self.p})'
class RandomAffineTransform(DualTransform):
"""Affine transformation of the input image with randomly chosen parameters. Image shape remains unchanged.
Args:
angle_limit (Tuple[float] | float, optional): Intervals in degrees from which angles of
rotation for the spatial axes are chosen.
Must be either of: ``A``, ``(A1, A2)``, ``(A1, A2, A3)``, or ``(A_Z1, A_Z2, A_Y1, A_Y2, A_X1, A_X2)``.
If a float, equivalent to ``(-A, A, -A, A, -A, A)``.
If a tuple with 2 items, equivalent to ``(A1, A2, A1, A2, A1, A2)``.
If a tuple with 3 items, equivalent to ``(-A1, A1, -A2, A2, -A3, A3)``.
If a tuple with 6 items, angle of rotation is randomly chosen from an interval [A_a1, A_a2] for each
spatial axis.
Defaults to ``(15, 15, 15)``.
translation_limit (Tuple[float] | float | None, optional): Intervals from which the translation parameters
Must be either of: ``T``, ``(T1, T2)``, ``(T1, T2, T3)``, or ``(T_Z1, T_Z2, T_Y1, T_Y2, T_X1, T_X2)``.
If a float, equivalent to ``(2-T, T, 2-T, T, 2-T, T)``.
If a tuple with 2 items, equivalent to ``(T1, T2, T1, T2, T1, T2)``.
If a tuple with 3 items, equivalent to ``(2-T1, T1, 2-T2, T2, 2-T3, T3)``.
If a tuple with 6 items, the translation parameter is randomly chosen from an interval [T_a1, T_a2] for
each spatial axis.
Defaults to ``(0, 0, 0)``.
scaling_limit (Tuple[float] | float, optional): Intervals from which the scales for the spatial axes are chosen.
Must be either of: ``S``, ``(S1, S2)``, ``(S1, S2, S3)``, or ``(S_Z1, S_Z2, S_Y1, S_Y2, S_X1, S_X2)``.
If a float, equivalent to ``(1/S, S, 1/S, S, 1/S, S)``.
If a tuple with 2 items, equivalent to ``(S1, S2, S1, S2, S1, S2)``.
If a tuple with 3 items, equivalent to ``(1/S1, S1, 1/S2, S2, 1/S3, S3)``.
If a tuple with 6 items, the scale is randomly chosen from an interval [S_a1, S_a2] for
each spatial axis.
spacing (float | Tuple[float, float, float] | None, optional): Voxel spacing for individual spatial dimensions.
Must be either of: ``S``, ``(S1, S2, S3)``, or ``None``.
If ``None``, equivalent to ``(1, 1, 1)``.
If a float ``S``, equivalent to ``(S, S, S)``.
Otherwise, a scale for each spatial dimension must be given.
Defaults to ``None``.
change_to_isotropic (bool, optional): Change data from anisotropic to isotropic.
Defaults to ``False``.
interpolation (str, optional): SimpleITK interpolation type for `image` and `float_mask`.
Must be one of ``linear``, ``nearest``, ``bspline``, ``gaussian``.
For `mask`, the ``nearest`` interpolation is always used.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'constant'``.
ival (float, optional): Value of `image` voxels outside of the `image` domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
mval (float, optional): Value of `mask` and `float_mask` voxels outside of the domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
ignore_index (float | None, optional): If a float, then transformation of `mask` is done with
``border_mode = 'constant'`` and ``mval = ignore_index``.
If ``None``, this argument is ignored.
Defaults to ``None``.
always_apply (bool, optional): Always apply this transformation in composition.
Defaults to ``False``.
p (float, optional): Chance of applying this transformation in composition.
Defaults to ``0.5``.
Targets:
image, mask, float mask, key points, bounding boxes
def __init__(self, angle_limit: Union[float, TypePairFloat, TypeTripletFloat, TypeSextetFloat] = (15., 15., 15.),
translation_limit: Union[float, TypePairFloat, TypeTripletFloat, TypeSextetFloat] = (0., 0., 0.),
scaling_limit: Union[float, TypePairFloat, TypeTripletFloat, TypeSextetFloat] = (1., 1., 1.),
spacing: Union[float, TypeTripletFloat] = None,
change_to_isotropic: bool = False,
border_mode: str = 'constant', ival: float = 0, mval: float = 0,
ignore_index: Union[float, None] = None, always_apply: bool = False, p: float = 0.5):
super().__init__(always_apply, p)
self.angle_limit: TypeSextetFloat = parse_limits(angle_limit)
self.translation_limit: TypeSextetFloat = parse_limits(translation_limit)
self.scaling_limit: TypeSextetFloat = parse_limits(scaling_limit, scale=True)
self.spacing: TypeTripletFloat = parse_coefs(spacing, identity_element=1)
self.interpolation: int = parse_itk_interpolation(interpolation)
self.border_mode = border_mode # not used
self.mask_mode = border_mode # not used
self.ival = ival
self.mval = mval
self.keep_scale = not change_to_isotropic
if ignore_index is not None:
self.mask_mode = "constant"
self.mval = ignore_index
def apply(self, img, **params):
return F.affine(img,
scales=params["scale"],
degrees=params["angles"],
translation=params["translation"],
interpolation=self.interpolation,
border_mode=self.border_mode,
value=self.ival,
spacing=self.spacing)
def apply_to_mask(self, mask, **params):
interpolation = parse_itk_interpolation('nearest') # refers to 'sitkNearestNeighbor'
scales=params["scale"],
degrees=params["angles"],
translation=params["translation"],
interpolation=interpolation,
border_mode=self.mask_mode,
value=self.mval,
scales=params["scale"],
degrees=params["angles"],
translation=params["translation"],
interpolation=self.interpolation,
border_mode=self.mask_mode,
value=self.mval,
def apply_to_keypoints(self, keypoints, **params):
return F.affine_keypoints(keypoints,
scales=params["scale"],
degrees=params["angles"],
translation=params["translation"],
spacing=self.spacing,
domain_limit=params['domain_limit'])
def get_params(self, **data):
# set parameters of the transform
scales = sample_range_uniform(self.scaling_limit)
angles = sample_range_uniform(self.angle_limit)
translation = sample_range_uniform(self.translation_limit)
domain_limit = get_spatio_temporal_domain_limit(data)
return {
"scale": scales,
"angles": angles,
"translation": translation,
"domain_limit": domain_limit
}
class AffineTransform(DualTransform):
"""Affine transformation of the input image with given parameters. Image shape remains unchanged.
Args:
angles (Tuple[float], optional): Angles of rotation for the spatial axes.
Must be: ``(A_Z, A_Y, A_X)``.
Defaults to ``(0, 0, 0)``.
translation (Tuple[float], optional): Translation vector for the spatial axes.
Must be: ``(T_Z, T_Y, T_X)``.
Defaults to ``(0, 0, 0)``.
scale (Tuple[float], optional): Scales for the spatial axes.
Must be: ``(S_Z, S_Y, S_X)``.
Defaults to ``(1, 1, 1)``.
spacing (Tuple[float, float, float], optional): Voxel spacing for individual spatial dimensions.
Must be: ``(S1, S2, S3)`` (a scale for each spatial dimension must be given).
Defaults to ``(1, 1, 1)``.
change_to_isotropic (bool, optional): Change data from anisotropic to isotropic.
Defaults to ``False``.
interpolation (str, optional): SimpleITK interpolation type for `image` and `float_mask`.
Must be one of ``linear``, ``nearest``, ``bspline``, ``gaussian``.
For `mask`, the ``nearest`` interpolation is always used.
border_mode (str, optional): Values outside image domain are filled according to this mode.
Defaults to ``'constant'``.
ival (float, optional): Value of `image` voxels outside of the `image` domain. Only applied when ``border_mode = 'constant'``.
Defaults to ``0``.
mval (float, optional): Value of `mask` and `float_mask` voxels outside of the domain. Only applied when ``border_mode = 'constant'``.