deepreg.dataset.preprocess.RandomDDFTransform3D.__init__() - Code Metrics - Inspection of "WIP 640 refactor model layers" - DeepRegNet/DeepReg - Measure and Improve Code Quality continuously with Scrutinizer

Passed

Pull Request — main (#656)

by Yunguan

created 2021-01-30 19:51 UTC

RandomDDFTransform3D.init() A

↳ Parent: deepreg.dataset.preprocess

Complexity

Conditions

Size

Total Lines	41
Code Lines	19

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	19
dl	0
loc	41
rs	9.45
c	0
b	0
f	0
cc	1
nop	8

How to fix Many Parameters

"""
Module containing data augmentation techniques.
  - 3D Affine/DDF Transforms for moving and fixed images.
"""

from abc import abstractmethod
from typing import Dict

import numpy as np
import tensorflow as tf

from deepreg.model.layer_util import get_reference_grid, resample, resize3d, warp_grid
from deepreg.registry import REGISTRY


class RandomTransformation3D(tf.keras.layers.Layer):
    """
    An interface for different types of transformation.
    """

    def __init__(
        self,
        moving_image_size: tuple,
        fixed_image_size: tuple,
        batch_size: int,
        name: str = "RandomTransformation3D",
        trainable: bool = False,
    ):
        """
        Abstract class for image transformation.

        :param moving_image_size: (m_dim1, m_dim2, m_dim3)
        :param fixed_image_size: (f_dim1, f_dim2, f_dim3)
        :param batch_size: size of mini-batch
        :param name: name of layer
        :param trainable: if this layer is trainable
        """
        super().__init__(trainable=trainable, name=name)
        self.moving_image_size = moving_image_size
        self.fixed_image_size = fixed_image_size
        self.batch_size = batch_size
        self.moving_grid_ref = get_reference_grid(grid_size=moving_image_size)
        self.fixed_grid_ref = get_reference_grid(grid_size=fixed_image_size)

    @abstractmethod
    def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
        """
        Generates transformation parameters for moving and fixed image.

        :return: two tensors
        """

    @staticmethod
    @abstractmethod
    def transform(
        image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
    ) -> tf.Tensor:
        """
        Transforms the reference grid and then resample the image.

        :param image: shape = (batch, dim1, dim2, dim3)
        :param grid_ref: shape = (dim1, dim2, dim3, 3)
        :param params: parameters for transformation
        :return: shape = (batch, dim1, dim2, dim3)
        """

    def call(self, inputs: Dict[str, tf.Tensor], **kwargs) -> Dict[str, tf.Tensor]:
        """
        Creates random params for the input images and their labels,
        and params them based on the resampled reference grids.
        :param inputs: a dict having multiple tensors
            if labeled:
                moving_image, shape = (batch, m_dim1, m_dim2, m_dim3)
                fixed_image, shape = (batch, f_dim1, f_dim2, f_dim3)
                moving_label, shape = (batch, m_dim1, m_dim2, m_dim3)
                fixed_label, shape = (batch, f_dim1, f_dim2, f_dim3)
                indices, shape = (batch, num_indices)
            else, unlabeled:
                moving_image, shape = (batch, m_dim1, m_dim2, m_dim3)
                fixed_image, shape = (batch, f_dim1, f_dim2, f_dim3)
                indices, shape = (batch, num_indices)
        :param kwargs: other arguments
        :return: dictionary with the same structure as inputs
        """

        moving_image = inputs["moving_image"]
        fixed_image = inputs["fixed_image"]
        indices = inputs["indices"]

        moving_params, fixed_params = self.gen_transform_params()

        moving_image = self.transform(moving_image, self.moving_grid_ref, moving_params)
        fixed_image = self.transform(fixed_image, self.fixed_grid_ref, fixed_params)

        if "moving_label" not in inputs:  # unlabeled
            return dict(
                moving_image=moving_image, fixed_image=fixed_image, indices=indices
            )
        moving_label = inputs["moving_label"]
        fixed_label = inputs["fixed_label"]

        moving_label = self.transform(moving_label, self.moving_grid_ref, moving_params)
        fixed_label = self.transform(fixed_label, self.fixed_grid_ref, fixed_params)

        return dict(
            moving_image=moving_image,
            fixed_image=fixed_image,
            moving_label=moving_label,
            fixed_label=fixed_label,
            indices=indices,
        )

    def get_config(self) -> dict:
        """Return the config dictionary for recreating this class."""
        config = super().get_config()
        config["moving_image_size"] = self.moving_image_size
        config["fixed_image_size"] = self.fixed_image_size
        config["batch_size"] = self.batch_size
        return config


@REGISTRY.register_data_augmentation(name="affine")
class RandomAffineTransform3D(RandomTransformation3D):
    """Apply random affine transformation to moving/fixed images separately."""

    def __init__(
        self,
        moving_image_size: tuple,
        fixed_image_size: tuple,
        batch_size: int,
        scale: float = 0.1,
        name: str = "RandomAffineTransform3D",
        **kwargs,
    ):
        """
        Init.

        :param moving_image_size: (m_dim1, m_dim2, m_dim3)
        :param fixed_image_size: (f_dim1, f_dim2, f_dim3)
        :param batch_size: size of mini-batch
        :param scale: a positive float controlling the scale of transformation
        :param name: name of the layer
        :param kwargs: extra arguments
        """
        super().__init__(
            moving_image_size=moving_image_size,
            fixed_image_size=fixed_image_size,
            batch_size=batch_size,
            name=name,
            **kwargs,
        )
        self.scale = scale

    def get_config(self) -> dict:
        """Return the config dictionary for recreating this class."""
        config = super().get_config()
        config["scale"] = self.scale
        return config

    def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
        """
        Function that generates the random 3D transformation parameters
        for a batch of data for moving and fixed image.

        :return: a tuple of tensors, each has shape = (batch, 4, 3)
        """
        theta = gen_rand_affine_transform(
            batch_size=self.batch_size * 2, scale=self.scale
        )
        return theta[: self.batch_size], theta[self.batch_size :]

    @staticmethod
    def transform(
        image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
    ) -> tf.Tensor:
        """
        Transforms the reference grid and then resample the image.

        :param image: shape = (batch, dim1, dim2, dim3)
        :param grid_ref: shape = (dim1, dim2, dim3, 3)
        :param params: shape = (batch, 4, 3)
        :return: shape = (batch, dim1, dim2, dim3)
        """
        return resample(vol=image, loc=warp_grid(grid_ref, params))


@REGISTRY.register_data_augmentation(name="ddf")
class RandomDDFTransform3D(RandomTransformation3D):
    """Apply random DDF transformation to moving/fixed images separately."""

    def __init__(
        self,
        moving_image_size: tuple,
        fixed_image_size: tuple,
        batch_size: int,
        field_strength: int = 1,
        low_res_size: tuple = (1, 1, 1),
        name: str = "RandomDDFTransform3D",
        **kwargs,
    ):
        """
        Creates a DDF transformation for data augmentation.

        To simulate smooth deformation fields, we interpolate from a low resolution
        field of size low_res_size using linear interpolation. The variance of the
        deformation field is drawn from a uniform variable
        between [0, field_strength].

        :param moving_image_size: tuple
        :param fixed_image_size: tuple
        :param batch_size: int
        :param field_strength: int = 1. It is used as the upper bound for the
        deformation field variance
        :param low_res_size: tuple = (1, 1, 1).
        :param name: name of layer
        :param kwargs: extra arguments
        """

        super().__init__(
            moving_image_size=moving_image_size,
            fixed_image_size=fixed_image_size,
            batch_size=batch_size,
            name=name,
            **kwargs,
        )

        assert tuple(low_res_size) <= tuple(moving_image_size)
        assert tuple(low_res_size) <= tuple(fixed_image_size)

        self.field_strength = field_strength
        self.low_res_size = low_res_size

    def get_config(self) -> dict:
        """Return the config dictionary for recreating this class."""
        config = super().get_config()
        config["field_strength"] = self.field_strength
        config["low_res_size"] = self.low_res_size
        return config

    def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
        """
        Generates two random ddf fields for moving and fixed images.

        :return: tuple, one has shape = (batch, m_dim1, m_dim2, m_dim3, 3)
            another one has shape = (batch, f_dim1, f_dim2, f_dim3, 3)
        """
        kwargs = dict(
            batch_size=self.batch_size,
            field_strength=self.field_strength,
            low_res_size=self.low_res_size,
        )
        moving = gen_rand_ddf(image_size=self.moving_image_size, **kwargs)
        fixed = gen_rand_ddf(image_size=self.fixed_image_size, **kwargs)
        return moving, fixed

    @staticmethod
    def transform(
        image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
    ) -> tf.Tensor:
        """
        Transforms the reference grid and then resample the image.

        :param image: shape = (batch, dim1, dim2, dim3)
        :param grid_ref: shape = (dim1, dim2, dim3, 3)
        :param params: DDF, shape = (batch, dim1, dim2, dim3, 3)
        :return: shape = (batch, dim1, dim2, dim3)
        """
        return resample(vol=image, loc=grid_ref[None, ...] + params)


def resize_inputs(
    inputs: Dict[str, tf.Tensor], moving_image_size: tuple, fixed_image_size: tuple
) -> Dict[str, tf.Tensor]:
    """
    Resize inputs
    :param inputs:
        if labeled:
            moving_image, shape = (None, None, None)
            fixed_image, shape = (None, None, None)
            moving_label, shape = (None, None, None)
            fixed_label, shape = (None, None, None)
            indices, shape = (num_indices, )
        else, unlabeled:
            moving_image, shape = (None, None, None)
            fixed_image, shape = (None, None, None)
            indices, shape = (num_indices, )
    :param moving_image_size: tuple, (m_dim1, m_dim2, m_dim3)
    :param fixed_image_size: tuple, (f_dim1, f_dim2, f_dim3)
    :return:
        if labeled:
            moving_image, shape = (m_dim1, m_dim2, m_dim3)
            fixed_image, shape = (f_dim1, f_dim2, f_dim3)
            moving_label, shape = (m_dim1, m_dim2, m_dim3)
            fixed_label, shape = (f_dim1, f_dim2, f_dim3)
            indices, shape = (num_indices, )
        else, unlabeled:
            moving_image, shape = (m_dim1, m_dim2, m_dim3)
            fixed_image, shape = (f_dim1, f_dim2, f_dim3)
            indices, shape = (num_indices, )
    """
    moving_image = inputs["moving_image"]
    fixed_image = inputs["fixed_image"]
    indices = inputs["indices"]

    moving_image = resize3d(image=moving_image, size=moving_image_size)
    fixed_image = resize3d(image=fixed_image, size=fixed_image_size)

    if "moving_label" not in inputs:  # unlabeled
        return dict(moving_image=moving_image, fixed_image=fixed_image, indices=indices)
    moving_label = inputs["moving_label"]
    fixed_label = inputs["fixed_label"]

    moving_label = resize3d(image=moving_label, size=moving_image_size)
    fixed_label = resize3d(image=fixed_label, size=fixed_image_size)

    return dict(
        moving_image=moving_image,
        fixed_image=fixed_image,
        moving_label=moving_label,
        fixed_label=fixed_label,
        indices=indices,
    )


def gen_rand_affine_transform(
    batch_size: int, scale: float, seed: (int, None) = None
) -> tf.Tensor:
    """
    Function that generates a random 3D transformation parameters for a batch of data.

    for 3D coordinates, affine transformation is

    .. code-block:: text

        [[x' y' z' 1]] = [[x y z 1]] * [[* * * 0]
                                        [* * * 0]
                                        [* * * 0]
                                        [* * * 1]]

    where each * represents a degree of freedom,
    so there are in total 12 degrees of freedom
    the equation can be denoted as

        new = old * T

    where

    - new is the transformed coordinates, of shape (1, 4)
    - old is the original coordinates, of shape (1, 4)
    - T is the transformation matrix, of shape (4, 4)

    the equation can be simplified to

    .. code-block:: text

        [[x' y' z']] = [[x y z 1]] * [[* * *]
                                      [* * *]
                                      [* * *]
                                      [* * *]]

    so that

        new = old * T

    where

    - new is the transformed coordinates, of shape (1, 3)
    - old is the original coordinates, of shape (1, 4)
    - T is the transformation matrix, of shape (4, 3)

    Given original and transformed coordinates,
    we can calculate the transformation matrix using

        x = np.linalg.lstsq(a, b)

    such that

        a x = b

    In our case,

    - a = old
    - b = new
    - x = T

    To generate random transformation,
    we choose to add random perturbation to corner coordinates as follows:
    for corner of coordinates (x, y, z), the noise is

        -(x, y, z) .* (r1, r2, r3)

    where ri is a random number between (0, scale).
    So

        (x', y', z') = (x, y, z) .* (1-r1, 1-r2, 1-r3)

    Thus, we can directly sample between 1-scale and 1 instead

    We choose to calculate the transformation based on
    four corners in a cube centered at (0, 0, 0).
    A cube is shown as below, where

    - C = (-1, -1, -1)
    - G = (-1, -1, 1)
    - D = (-1, 1, -1)
    - A = (1, -1, -1)

    .. code-block:: text

                    G — — — — — — — — H
                  / |               / |
                /   |             /   |
              /     |           /     |
            /       |         /       |
          /         |       /         |
        E — — — — — — — — F           |
        |           |     |           |
        |           |     |           |
        |           C — — | — — — — — D
        |         /       |         /
        |       /         |       /
        |     /           |     /
        |   /             |   /
        | /               | /
        A — — — — — — — — B

    :param batch_size: int
    :param scale: a float number between 0 and 1
    :param seed: control the randomness
    :return: shape = (batch, 4, 3)
    """

    assert 0 <= scale <= 1
    np.random.seed(seed)
    noise = np.random.uniform(1 - scale, 1, [batch_size, 4, 3])  # shape = (batch, 4, 3)

    # old represents four corners of a cube
    # corresponding to the corner C G D A as shown above
    old = np.tile(
        [[[-1, -1, -1, 1], [-1, -1, 1, 1], [-1, 1, -1, 1], [1, -1, -1, 1]]],
        [batch_size, 1, 1],
    )  # shape = (batch, 4, 4)
    new = old[:, :, :3] * noise  # shape = (batch, 4, 3)

    theta = np.array(
        [np.linalg.lstsq(old[k], new[k], rcond=-1)[0] for k in range(batch_size)]
    )  # shape = (batch, 4, 3)

    return tf.cast(theta, dtype=tf.float32)


def gen_rand_ddf(
    batch_size: int,
    image_size: tuple,
    field_strength: (tuple, list),
    low_res_size: (tuple, list),
    seed: (int, None) = None,
) -> tf.Tensor:
    """
    Function that generates a random 3D DDF for a batch of data.

    :param batch_size:
    :param image_size:
    :param field_strength: maximum field strength, computed as a U[0,field_strength]
    :param low_res_size: low_resolution deformation field that will be upsampled to
        the original size in order to get smooth and more realistic fields.
    :param seed: control the randomness
    :return:
    """

    np.random.seed(seed)
    low_res_strength = np.random.uniform(0, field_strength, (batch_size, 1, 1, 1, 3))
    low_res_field = low_res_strength * np.random.randn(
        batch_size, low_res_size[0], low_res_size[1], low_res_size[2], 3
    )
    high_res_field = resize3d(low_res_field, image_size)
    return high_res_field


1			"""
2			Module containing data augmentation techniques.
3			- 3D Affine/DDF Transforms for moving and fixed images.
4			"""
5
6			from abc import abstractmethod
7			from typing import Dict
8
9			import numpy as np
10			import tensorflow as tf
11
12			from deepreg.model.layer_util import get_reference_grid, resample, resize3d, warp_grid
13			from deepreg.registry import REGISTRY
14
15
16			class RandomTransformation3D(tf.keras.layers.Layer):
17			"""
18			An interface for different types of transformation.
19			"""
20
21			def __init__(
22			self,
23			moving_image_size: tuple,
24			fixed_image_size: tuple,
25			batch_size: int,
26			name: str = "RandomTransformation3D",
27			trainable: bool = False,
28			):
29			"""
30			Abstract class for image transformation.
31
32			:param moving_image_size: (m_dim1, m_dim2, m_dim3)
33			:param fixed_image_size: (f_dim1, f_dim2, f_dim3)
34			:param batch_size: size of mini-batch
35			:param name: name of layer
36			:param trainable: if this layer is trainable
37			"""
38			super().__init__(trainable=trainable, name=name)
39			self.moving_image_size = moving_image_size
40			self.fixed_image_size = fixed_image_size
41			self.batch_size = batch_size
42			self.moving_grid_ref = get_reference_grid(grid_size=moving_image_size)
43			self.fixed_grid_ref = get_reference_grid(grid_size=fixed_image_size)
44
45			@abstractmethod
46			def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
47			"""
48			Generates transformation parameters for moving and fixed image.
49
50			:return: two tensors
51			"""
52
53			@staticmethod
54			@abstractmethod
55			def transform(
56			image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
57			) -> tf.Tensor:
58			"""
59			Transforms the reference grid and then resample the image.
60
61			:param image: shape = (batch, dim1, dim2, dim3)
62			:param grid_ref: shape = (dim1, dim2, dim3, 3)
63			:param params: parameters for transformation
64			:return: shape = (batch, dim1, dim2, dim3)
65			"""
66
67			def call(self, inputs: Dict[str, tf.Tensor], **kwargs) -> Dict[str, tf.Tensor]:
68			"""
69			Creates random params for the input images and their labels,
70			and params them based on the resampled reference grids.
71			:param inputs: a dict having multiple tensors
72			if labeled:
73			moving_image, shape = (batch, m_dim1, m_dim2, m_dim3)
74			fixed_image, shape = (batch, f_dim1, f_dim2, f_dim3)
75			moving_label, shape = (batch, m_dim1, m_dim2, m_dim3)
76			fixed_label, shape = (batch, f_dim1, f_dim2, f_dim3)
77			indices, shape = (batch, num_indices)
78			else, unlabeled:
79			moving_image, shape = (batch, m_dim1, m_dim2, m_dim3)
80			fixed_image, shape = (batch, f_dim1, f_dim2, f_dim3)
81			indices, shape = (batch, num_indices)
82			:param kwargs: other arguments
83			:return: dictionary with the same structure as inputs
84			"""
85
86			moving_image = inputs["moving_image"]
87			fixed_image = inputs["fixed_image"]
88			indices = inputs["indices"]
89
90			moving_params, fixed_params = self.gen_transform_params()
91
92			moving_image = self.transform(moving_image, self.moving_grid_ref, moving_params)
93			fixed_image = self.transform(fixed_image, self.fixed_grid_ref, fixed_params)
94
95			if "moving_label" not in inputs: # unlabeled
96			return dict(
97			moving_image=moving_image, fixed_image=fixed_image, indices=indices
98			)
99			moving_label = inputs["moving_label"]
100			fixed_label = inputs["fixed_label"]
101
102			moving_label = self.transform(moving_label, self.moving_grid_ref, moving_params)
103			fixed_label = self.transform(fixed_label, self.fixed_grid_ref, fixed_params)
104
105			return dict(
106			moving_image=moving_image,
107			fixed_image=fixed_image,
108			moving_label=moving_label,
109			fixed_label=fixed_label,
110			indices=indices,
111			)
112
113			def get_config(self) -> dict:
114			"""Return the config dictionary for recreating this class."""
115			config = super().get_config()
116			config["moving_image_size"] = self.moving_image_size
117			config["fixed_image_size"] = self.fixed_image_size
118			config["batch_size"] = self.batch_size
119			return config
120
121
122			@REGISTRY.register_data_augmentation(name="affine")
123			class RandomAffineTransform3D(RandomTransformation3D):
124			"""Apply random affine transformation to moving/fixed images separately."""
125
126			def __init__(
127			self,
128			moving_image_size: tuple,
129			fixed_image_size: tuple,
130			batch_size: int,
131			scale: float = 0.1,
132			name: str = "RandomAffineTransform3D",
133			**kwargs,
134			):
135			"""
136			Init.
137
138			:param moving_image_size: (m_dim1, m_dim2, m_dim3)
139			:param fixed_image_size: (f_dim1, f_dim2, f_dim3)
140			:param batch_size: size of mini-batch
141			:param scale: a positive float controlling the scale of transformation
142			:param name: name of the layer
143			:param kwargs: extra arguments
144			"""
145			super().__init__(
146			moving_image_size=moving_image_size,
147			fixed_image_size=fixed_image_size,
148			batch_size=batch_size,
149			name=name,
150			**kwargs,
151			)
152			self.scale = scale
153
154			def get_config(self) -> dict:
155			"""Return the config dictionary for recreating this class."""
156			config = super().get_config()
157			config["scale"] = self.scale
158			return config
159
160			def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
161			"""
162			Function that generates the random 3D transformation parameters
163			for a batch of data for moving and fixed image.
164
165			:return: a tuple of tensors, each has shape = (batch, 4, 3)
166			"""
167			theta = gen_rand_affine_transform(
168			batch_size=self.batch_size * 2, scale=self.scale
169			)
170			return theta[: self.batch_size], theta[self.batch_size :]
171
172			@staticmethod
173			def transform(
174			image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
175			) -> tf.Tensor:
176			"""
177			Transforms the reference grid and then resample the image.
178
179			:param image: shape = (batch, dim1, dim2, dim3)
180			:param grid_ref: shape = (dim1, dim2, dim3, 3)
181			:param params: shape = (batch, 4, 3)
182			:return: shape = (batch, dim1, dim2, dim3)
183			"""
184			return resample(vol=image, loc=warp_grid(grid_ref, params))
185
186
187			@REGISTRY.register_data_augmentation(name="ddf")
188			class RandomDDFTransform3D(RandomTransformation3D):
189			"""Apply random DDF transformation to moving/fixed images separately."""
190
191			def __init__(
192			self,
193			moving_image_size: tuple,
194			fixed_image_size: tuple,
195			batch_size: int,
196			field_strength: int = 1,
197			low_res_size: tuple = (1, 1, 1),
198			name: str = "RandomDDFTransform3D",
199			**kwargs,
200			):
201			"""
202			Creates a DDF transformation for data augmentation.
203
204			To simulate smooth deformation fields, we interpolate from a low resolution
205			field of size low_res_size using linear interpolation. The variance of the
206			deformation field is drawn from a uniform variable
207			between [0, field_strength].
208
209			:param moving_image_size: tuple
210			:param fixed_image_size: tuple
211			:param batch_size: int
212			:param field_strength: int = 1. It is used as the upper bound for the
213			deformation field variance
214			:param low_res_size: tuple = (1, 1, 1).
215			:param name: name of layer
216			:param kwargs: extra arguments
217			"""
218
219			super().__init__(
220			moving_image_size=moving_image_size,
221			fixed_image_size=fixed_image_size,
222			batch_size=batch_size,
223			name=name,
224			**kwargs,
225			)
226
227			assert tuple(low_res_size) <= tuple(moving_image_size)
228			assert tuple(low_res_size) <= tuple(fixed_image_size)
229
230			self.field_strength = field_strength
231			self.low_res_size = low_res_size
232
233			def get_config(self) -> dict:
234			"""Return the config dictionary for recreating this class."""
235			config = super().get_config()
236			config["field_strength"] = self.field_strength
237			config["low_res_size"] = self.low_res_size
238			return config
239
240			def gen_transform_params(self) -> (tf.Tensor, tf.Tensor):
241			"""
242			Generates two random ddf fields for moving and fixed images.
243
244			:return: tuple, one has shape = (batch, m_dim1, m_dim2, m_dim3, 3)
245			another one has shape = (batch, f_dim1, f_dim2, f_dim3, 3)
246			"""
247			kwargs = dict(
248			batch_size=self.batch_size,
249			field_strength=self.field_strength,
250			low_res_size=self.low_res_size,
251			)
252			moving = gen_rand_ddf(image_size=self.moving_image_size, **kwargs)
253			fixed = gen_rand_ddf(image_size=self.fixed_image_size, **kwargs)
254			return moving, fixed
255
256			@staticmethod
257			def transform(
258			image: tf.Tensor, grid_ref: tf.Tensor, params: tf.Tensor
259			) -> tf.Tensor:
260			"""
261			Transforms the reference grid and then resample the image.
262
263			:param image: shape = (batch, dim1, dim2, dim3)
264			:param grid_ref: shape = (dim1, dim2, dim3, 3)
265			:param params: DDF, shape = (batch, dim1, dim2, dim3, 3)
266			:return: shape = (batch, dim1, dim2, dim3)
267			"""
268			return resample(vol=image, loc=grid_ref[None, ...] + params)
269
270
271			def resize_inputs(
272			inputs: Dict[str, tf.Tensor], moving_image_size: tuple, fixed_image_size: tuple
273			) -> Dict[str, tf.Tensor]:
274			"""
275			Resize inputs
276			:param inputs:
277			if labeled:
278			moving_image, shape = (None, None, None)
279			fixed_image, shape = (None, None, None)
280			moving_label, shape = (None, None, None)
281			fixed_label, shape = (None, None, None)
282			indices, shape = (num_indices, )
283			else, unlabeled:
284			moving_image, shape = (None, None, None)
285			fixed_image, shape = (None, None, None)
286			indices, shape = (num_indices, )
287			:param moving_image_size: tuple, (m_dim1, m_dim2, m_dim3)
288			:param fixed_image_size: tuple, (f_dim1, f_dim2, f_dim3)
289			:return:
290			if labeled:
291			moving_image, shape = (m_dim1, m_dim2, m_dim3)
292			fixed_image, shape = (f_dim1, f_dim2, f_dim3)
293			moving_label, shape = (m_dim1, m_dim2, m_dim3)
294			fixed_label, shape = (f_dim1, f_dim2, f_dim3)
295			indices, shape = (num_indices, )
296			else, unlabeled:
297			moving_image, shape = (m_dim1, m_dim2, m_dim3)
298			fixed_image, shape = (f_dim1, f_dim2, f_dim3)
299			indices, shape = (num_indices, )
300			"""
301			moving_image = inputs["moving_image"]
302			fixed_image = inputs["fixed_image"]
303			indices = inputs["indices"]
304
305			moving_image = resize3d(image=moving_image, size=moving_image_size)
306			fixed_image = resize3d(image=fixed_image, size=fixed_image_size)
307
308			if "moving_label" not in inputs: # unlabeled
309			return dict(moving_image=moving_image, fixed_image=fixed_image, indices=indices)
310			moving_label = inputs["moving_label"]
311			fixed_label = inputs["fixed_label"]
312
313			moving_label = resize3d(image=moving_label, size=moving_image_size)
314			fixed_label = resize3d(image=fixed_label, size=fixed_image_size)
315
316			return dict(
317			moving_image=moving_image,
318			fixed_image=fixed_image,
319			moving_label=moving_label,
320			fixed_label=fixed_label,
321			indices=indices,
322			)
323
324
325			def gen_rand_affine_transform(
326			batch_size: int, scale: float, seed: (int, None) = None
327			) -> tf.Tensor:
328			"""
329			Function that generates a random 3D transformation parameters for a batch of data.
330
331			for 3D coordinates, affine transformation is
332
333			.. code-block:: text
334
335			[[x' y' z' 1]] = [[x y z 1]] * [[* * * 0]
336			[* * * 0]
337			[* * * 0]
338			[* * * 1]]
339
340			where each * represents a degree of freedom,
341			so there are in total 12 degrees of freedom
342			the equation can be denoted as
343
344			new = old * T
345
346			where
347
348			- new is the transformed coordinates, of shape (1, 4)
349			- old is the original coordinates, of shape (1, 4)
350			- T is the transformation matrix, of shape (4, 4)
351
352			the equation can be simplified to
353
354			.. code-block:: text
355
356			[[x' y' z']] = [[x y z 1]] * [[* * *]
357			[* * *]
358			[* * *]
359			[* * *]]
360
361			so that
362
363			new = old * T
364
365			where
366
367			- new is the transformed coordinates, of shape (1, 3)
368			- old is the original coordinates, of shape (1, 4)
369			- T is the transformation matrix, of shape (4, 3)
370
371			Given original and transformed coordinates,
372			we can calculate the transformation matrix using
373
374			x = np.linalg.lstsq(a, b)
375
376			such that
377
378			a x = b
379
380			In our case,
381
382			- a = old
383			- b = new
384			- x = T
385
386			To generate random transformation,
387			we choose to add random perturbation to corner coordinates as follows:
388			for corner of coordinates (x, y, z), the noise is
389
390			-(x, y, z) .* (r1, r2, r3)
391
392			where ri is a random number between (0, scale).
393			So
394
395			(x', y', z') = (x, y, z) .* (1-r1, 1-r2, 1-r3)
396
397			Thus, we can directly sample between 1-scale and 1 instead
398
399			We choose to calculate the transformation based on
400			four corners in a cube centered at (0, 0, 0).
401			A cube is shown as below, where
402
403			- C = (-1, -1, -1)
404			- G = (-1, -1, 1)
405			- D = (-1, 1, -1)
406			- A = (1, -1, -1)
407
408			.. code-block:: text
409
410			G — — — — — — — — H
411			/ \| / \|
412			/ \| / \|
413			/ \| / \|
414			/ \| / \|
415			/ \| / \|
416			E — — — — — — — — F \|
417			\| \| \| \|
418			\| \| \| \|
419			\| C — — \| — — — — — D
420			\| / \| /
421			\| / \| /
422			\| / \| /
423			\| / \| /
424			\| / \| /
425			A — — — — — — — — B
426
427			:param batch_size: int
428			:param scale: a float number between 0 and 1
429			:param seed: control the randomness
430			:return: shape = (batch, 4, 3)
431			"""
432
433			assert 0 <= scale <= 1
434			np.random.seed(seed)
435			noise = np.random.uniform(1 - scale, 1, [batch_size, 4, 3]) # shape = (batch, 4, 3)
436
437			# old represents four corners of a cube
438			# corresponding to the corner C G D A as shown above
439			old = np.tile(
440			[[[-1, -1, -1, 1], [-1, -1, 1, 1], [-1, 1, -1, 1], [1, -1, -1, 1]]],
441			[batch_size, 1, 1],
442			) # shape = (batch, 4, 4)
443			new = old[:, :, :3] * noise # shape = (batch, 4, 3)
444
445			theta = np.array(
446			[np.linalg.lstsq(old[k], new[k], rcond=-1)[0] for k in range(batch_size)]
447			) # shape = (batch, 4, 3)
448
449			return tf.cast(theta, dtype=tf.float32)
450
451
452			def gen_rand_ddf(
453			batch_size: int,
454			image_size: tuple,
455			field_strength: (tuple, list),
456			low_res_size: (tuple, list),
457			seed: (int, None) = None,
458			) -> tf.Tensor:
459			"""
460			Function that generates a random 3D DDF for a batch of data.
461
462			:param batch_size:
463			:param image_size:
464			:param field_strength: maximum field strength, computed as a U[0,field_strength]
465			:param low_res_size: low_resolution deformation field that will be upsampled to
466			the original size in order to get smooth and more realistic fields.
467			:param seed: control the randomness
468			:return:
469			"""
470
471			np.random.seed(seed)
472			low_res_strength = np.random.uniform(0, field_strength, (batch_size, 1, 1, 1, 3))
473			low_res_field = low_res_strength * np.random.randn(
474			batch_size, low_res_size[0], low_res_size[1], low_res_size[2], 3
475			)
476			high_res_field = resize3d(low_res_field, image_size)
477			return high_res_field
478

DeepRegNet / DeepReg

Pull Request — main (#656)

RandomDDFTransform3D.__init__() A

Complexity

Size

Duplication

Importance

How to fix Many Parameters

Many Parameters

Duplication Side-by-Side

Filter issues like

RandomDDFTransform3D.init() A