deepreg.model.backbone.u_net.AbstractUNet.build_layers() - 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 2021-02-17 22:31 UTC

AbstractUNet.build_layers() C

↳ Parent: deepreg.model.backbone.u_net

Complexity

Conditions

Size

Total Lines	111
Code Lines	77

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	77
dl	0
loc	111
rs	6.8193
c	0
b	0
f	0
cc	6
nop	12

How to fix Long Method Many Parameters

# coding=utf-8



from abc import abstractmethod
from typing import List, Tuple, Union

import tensorflow as tf
import tensorflow.keras.layers as tfkl
from tensorflow.python.keras.utils import conv_utils

from deepreg.model import layer, layer_util
from deepreg.model.backbone.interface import Backbone
from deepreg.registry import REGISTRY


class AbstractUNet(Backbone):
    """An interface for u-net style backbones."""

    def __init__(

        self,
        image_size: tuple,
        num_channel_initial: int,
        depth: int,
        extract_levels: List[int],
        out_kernel_initializer: str,
        out_activation: str,
        out_channels: int,
        name: str = "AbstractUNet",
        **kwargs,
    ):
        """
        Init.

        Image is encoded gradually, i from level 0 to D,
        then it is decoded gradually, j from level D to 0.
        Some of the decoded levels are used for generating extractions.

        So, extract_levels are between [0, D].

        :param image_size: such as (dim1, dim2, dim3)
        :param num_channel_initial: number of initial channels.
        :param depth: d = 0 to depth, both side included
        :param extract_levels: from which depths the output will be built.
        :param out_kernel_initializer: initializer to use for kernels.
        :param out_activation: activation to use at end layer.
        :param out_channels: number of channels for the extractions
        :param use_additive_upsampling: whether use additive up-sampling.
        :param name: name of the backbone.
        :param kwargs: additional arguments.
        """
        super().__init__(
            image_size=image_size,
            out_channels=out_channels,
            num_channel_initial=num_channel_initial,
            out_kernel_initializer=out_kernel_initializer,
            out_activation=out_activation,
            name=name,
            **kwargs,
        )

        # save parameters
        assert max(extract_levels) <= depth
        self._extract_levels = extract_levels
        self._depth = depth

        # init layers
        # all lists start with d = 0
        self._downsample_convs = None
        self._downsample_pools = None
        self._bottom_block = None
        self._upsample_deconvs = None
        self._upsample_convs = None
        self._output_block = None

    def build_layers(
        self,
        image_size: tuple,
        num_channel_initial: int,
        depth: int,
        extract_levels: List[int],
        downsample_kernel_sizes: Union[int, List[int]],
        upsample_kernel_sizes: Union[int, List[int]],
        strides: int,
        padding: str,
        out_kernel_initializer: str,
        out_activation: str,
        out_channels: int,
    ):
        """
        Build layers that will be used in call.

        :param image_size: (dim1, dim2, dim3).
        :param num_channel_initial: number of initial channels.
        :param depth: network starts with d = 0, and the bottom has d = depth.
        :param extract_levels: from which depths the output will be built.
        :param downsample_kernel_sizes: kernel size for down-sampling
        :param upsample_kernel_sizes: kernel size for up-sampling
        :param strides: strides for down-sampling
        :param padding: padding mode for all conv layers
        :param out_kernel_initializer: initializer to use for kernels.
        :param out_activation: activation to use at end layer.
        :param out_channels: number of channels for the extractions
        """
        # init params
        min_extract_level = min(extract_levels)
        num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]
        if isinstance(downsample_kernel_sizes, int):
            downsample_kernel_sizes = [downsample_kernel_sizes] * (depth + 1)
        assert len(downsample_kernel_sizes) == depth + 1
        if isinstance(upsample_kernel_sizes, int):
            upsample_kernel_sizes = [upsample_kernel_sizes] * depth
        assert len(upsample_kernel_sizes) == depth

        # down-sampling
        self._downsample_convs = []
        self._downsample_pools = []
        tensor_shape = image_size
        tensor_shapes = [tensor_shape]
        for d in range(depth):
            downsample_conv = self.build_conv_block(
                filters=num_channels[d],
                kernel_size=downsample_kernel_sizes[d],
                padding=padding,
            )
            downsample_pool = self.build_down_sampling_block(
                kernel_size=strides, strides=strides, padding=padding
            )
            tensor_shape = tuple(
                conv_utils.conv_output_length(
                    input_length=x,
                    filter_size=strides,
                    padding=padding,
                    stride=strides,
                    dilation=1,
                )
                for x in tensor_shape
            )
            self._downsample_convs.append(downsample_conv)
            self._downsample_pools.append(downsample_pool)
            tensor_shapes.append(tensor_shape)

        # bottom layer
        self._bottom_block = self.build_bottom_block(
            filters=num_channels[depth],
            kernel_size=downsample_kernel_sizes[depth],
            padding=padding,
        )

        # up-sampling
        self._upsample_deconvs = []
        self._upsample_convs = []
        for d in range(depth - 1, min_extract_level - 1, -1):
            kernel_size = upsample_kernel_sizes[d]
            output_padding = layer_util.deconv_output_padding(
                input_shape=tensor_shapes[d + 1],
                output_shape=tensor_shapes[d],
                kernel_size=kernel_size,
                stride=strides,
                padding=padding,
            )
            upsample_deconv = self.build_up_sampling_block(
                filters=num_channels[d],
                output_padding=output_padding,
                kernel_size=kernel_size,
                strides=strides,
                padding=padding,
                output_shape=tensor_shapes[d],
            )
            upsample_conv = self.build_conv_block(
                filters=num_channels[d], kernel_size=kernel_size, padding=padding
            )
            self._upsample_deconvs = [upsample_deconv] + self._upsample_deconvs
            self._upsample_convs = [upsample_conv] + self._upsample_convs
        if min_extract_level > 0:
            # add Nones to make lists have length depth - 1
            self._upsample_deconvs = [None] * min_extract_level + self._upsample_deconvs
            self._upsample_convs = [None] * min_extract_level + self._upsample_convs

        # extraction
        self._output_block = self.build_output_block(
            image_size=image_size,
            extract_levels=extract_levels,
            out_channels=out_channels,
            out_kernel_initializer=out_kernel_initializer,
            out_activation=out_activation,
        )

    @abstractmethod
    def build_conv_block(
        self, filters: int, kernel_size: int, padding: str
    ) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a conv block for down-sampling or up-sampling.

        This block do not change the tensor shape (width, height, depth),
        it only changes the number of channels.

        :param filters: number of channels for output
        :param kernel_size: arg for conv3d
        :param padding: arg for conv3d
        :return: a block consists of one or multiple layers
        """

    @abstractmethod
    def build_down_sampling_block(
        self, kernel_size: int, padding: str, strides: int
    ) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a block for down-sampling.

        This block changes the tensor shape (width, height, depth),
        but it does not changes the number of channels.

        :param kernel_size: arg for pool3d
        :param padding: arg for pool3d
        :param strides: arg for pool3d
        :return: a block consists of one or multiple layers
        """

    @abstractmethod
    def build_bottom_block(
        self, filters: int, kernel_size: int, padding: str
    ) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a block for bottom layer.

        This block do not change the tensor shape (width, height, depth),
        it only changes the number of channels.

        :param filters: number of channels for output
        :param kernel_size: arg for conv3d
        :param padding: arg for conv3d
        :return: a block consists of one or multiple layers
        """

    @abstractmethod
    def build_up_sampling_block(
        self,
        filters: int,
        output_padding: int,
        kernel_size: int,
        padding: str,
        strides: int,
        output_shape: tuple,
    ) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a block for up-sampling.

        This block changes the tensor shape (width, height, depth),
        but it does not changes the number of channels.

        :param filters: number of channels for output
        :param output_padding: padding for output
        :param kernel_size: arg for deconv3d
        :param padding: arg for deconv3d
        :param strides: arg for deconv3d
        :param output_shape: shape of the output tensor
        :return: a block consists of one or multiple layers
        """

    @abstractmethod
    def build_skip_block(self) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a block for combining skipped tensor and up-sampled one.

        This block do not change the tensor shape (width, height, depth),
        it only changes the number of channels.

        The input to this block is a list of tensors.

        :return: a block consists of one or multiple layers
        """

    @abstractmethod
    def build_output_block(
        self,
        image_size: Tuple[int],
        extract_levels: List[int],
        out_channels: int,
        out_kernel_initializer: str,
        out_activation: str,
    ) -> Union[tf.keras.Model, tfkl.Layer]:
        """
        Build a block for output.

        The input to this block is a list of tensors.

        :param image_size: such as (dim1, dim2, dim3)
        :param extract_levels: number of extraction levels.
        :param out_channels: number of channels for the extractions
        :param out_kernel_initializer: initializer to use for kernels.
        :param out_activation: activation to use at end layer.
        :return: a block consists of one or multiple layers
        """

    def call(self, inputs: tf.Tensor, training=None, mask=None) -> tf.Tensor:

        """
        Build LocalNet graph based on built layers.

        :param inputs: image batch, shape = (batch, f_dim1, f_dim2, f_dim3, ch)
        :param training: None or bool.
        :param mask: None or tf.Tensor.
        :return: shape = (batch, f_dim1, f_dim2, f_dim3, out_channels)
        """

        # down-sampling
        skips = []
        down_sampled = inputs
        for d in range(self._depth):
            skip = self._downsample_convs[d](inputs=down_sampled, training=training)
            down_sampled = self._downsample_pools[d](inputs=skip, training=training)
            skips.append(skip)

        # bottom
        up_sampled = self._bottom_block(inputs=down_sampled, training=training)

        # up-sampling
        outs = [up_sampled]
        for d in range(self._depth - 1, min(self._extract_levels) - 1, -1):
            up_sampled = self._upsample_deconvs[d](inputs=up_sampled, training=training)
            up_sampled = self.build_skip_block()([up_sampled, skips[d]])
            up_sampled = self._upsample_convs[d](inputs=up_sampled, training=training)
            outs.append(up_sampled)

        # output
        output = self._output_block(outs)

        return output


@REGISTRY.register_backbone(name="unet")
class UNet(Backbone):
    """
    Class that implements an adapted 3D UNet.

    Reference:

    - O. Ronneberger, P. Fischer, and T. Brox,
      “U-net: Convolutional networks for biomedical image segmentation,”,
      Lecture Notes in Computer Science, 2015, vol. 9351, pp. 234–241.
      https://arxiv.org/abs/1505.04597
    """

    def __init__(
        self,
        image_size: tuple,
        out_channels: int,
        num_channel_initial: int,
        depth: int,
        out_kernel_initializer: str,
        out_activation: str,
        pooling: bool = True,
        concat_skip: bool = False,
        name: str = "Unet",
        **kwargs,
    ):
        """
        Initialise UNet.

        :param image_size: (dim1, dim2, dim3), dims of input image.
        :param out_channels: number of channels for the output
        :param num_channel_initial: number of initial channels
        :param depth: input is at level 0, bottom is at level depth
        :param out_kernel_initializer: kernel initializer for the last layer
        :param out_activation: activation at the last layer
        :param pooling: for downsampling, use non-parameterized
                        pooling if true, otherwise use conv3d
        :param concat_skip: when upsampling, concatenate skipped
                            tensor if true, otherwise use addition
        :param name: name of the backbone.
        :param kwargs: additional arguments.
        """
        super().__init__(
            image_size=image_size,
            out_channels=out_channels,
            num_channel_initial=num_channel_initial,
            out_kernel_initializer=out_kernel_initializer,
            out_activation=out_activation,
            name=name,
            **kwargs,
        )

        # init layer variables
        num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]

        self._num_channel_initial = num_channel_initial
        self._depth = depth
        self._concat_skip = concat_skip
        self._downsample_convs = []
        self._downsample_pools = []
        tensor_shape = image_size
        self._tensor_shapes = [tensor_shape]
        for d in range(depth):
            downsample_conv = tf.keras.Sequential(
                [
                    layer.Conv3dBlock(
                        filters=num_channels[d], kernel_size=3, padding="same"
                    ),
                    layer.ResidualConv3dBlock(
                        filters=num_channels[d], kernel_size=3, padding="same"
                    ),
                ]
            )
            if pooling:
                downsample_pool = tfkl.MaxPool3D(pool_size=2, strides=2, padding="same")
            else:
                downsample_pool = layer.Conv3dBlock(
                    filters=num_channels[d], kernel_size=3, strides=2, padding="same"
                )
            tensor_shape = tuple((x + 1) // 2 for x in tensor_shape)
            self._downsample_convs.append(downsample_conv)
            self._downsample_pools.append(downsample_pool)
            self._tensor_shapes.append(tensor_shape)
        self._bottom_conv3d = layer.Conv3dBlock(
            filters=num_channels[depth], kernel_size=3, padding="same"
        )
        self._bottom_res3d = layer.ResidualConv3dBlock(
            filters=num_channels[depth], kernel_size=3, padding="same"
        )
        self._upsample_deconvs = []
        self._upsample_convs = []
        for d in range(depth):
            padding = layer_util.deconv_output_padding(
                input_shape=self._tensor_shapes[d + 1],
                output_shape=self._tensor_shapes[d],
                kernel_size=3,
                stride=2,
                padding="same",
            )
            upsample_deconv = layer.Deconv3dBlock(
                filters=num_channels[d],
                output_padding=padding,
                kernel_size=3,
                strides=2,
                padding="same",
            )
            upsample_conv = tf.keras.Sequential(
                [
                    layer.Conv3dBlock(
                        filters=num_channels[d], kernel_size=3, padding="same"
                    ),
                    layer.ResidualConv3dBlock(
                        filters=num_channels[d], kernel_size=3, padding="same"
                    ),
                ]
            )
            self._upsample_deconvs.append(upsample_deconv)
            self._upsample_convs.append(upsample_conv)
        self._output_conv3d = tf.keras.Sequential(
            [
                tfkl.Conv3D(
                    filters=out_channels,
                    kernel_size=3,
                    strides=1,
                    padding="same",
                    kernel_initializer=out_kernel_initializer,
                    activation=out_activation,
                ),
                layer.Resize3d(shape=image_size),
            ]
        )

    def call(self, inputs: tf.Tensor, training=None, mask=None) -> tf.Tensor:

        """
        Builds graph based on built layers.

        :param inputs: shape = [batch, f_dim1, f_dim2, f_dim3, in_channels]
        :param training:
        :param mask:
        :return: shape = [batch, f_dim1, f_dim2, f_dim3, out_channels]
        """

        down_sampled = inputs

        # down sample
        skips = []
        for d_var in range(self._depth):  # level 0 to D-1
            skip = self._downsample_convs[d_var](inputs=down_sampled, training=training)
            down_sampled = self._downsample_pools[d_var](inputs=skip, training=training)
            skips.append(skip)

        # bottom, level D
        up_sampled = self._bottom_res3d(
            inputs=self._bottom_conv3d(inputs=down_sampled, training=training),
            training=training,
        )

        # up sample, level D-1 to 0
        for d_var in range(self._depth - 1, -1, -1):
            up_sampled = self._upsample_deconvs[d_var](
                inputs=up_sampled, training=training
            )
            if self._concat_skip:
                up_sampled = tf.concat([up_sampled, skips[d_var]], axis=4)
            else:
                up_sampled = up_sampled + skips[d_var]
            up_sampled = self._upsample_convs[d_var](
                inputs=up_sampled, training=training
            )

        # output
        output = self._output_conv3d(inputs=up_sampled, training=training)

        return output


1			# coding=utf-8
			0 ignored issues – show introduced 2021-02-17 22:43 UTC by Report Bug Copy Issue Report Missing module docstring Loading history...
2
3
4			from abc import abstractmethod
5			from typing import List, Tuple, Union
6
7			import tensorflow as tf
8			import tensorflow.keras.layers as tfkl
9			from tensorflow.python.keras.utils import conv_utils
10
11			from deepreg.model import layer, layer_util
12			from deepreg.model.backbone.interface import Backbone
13			from deepreg.registry import REGISTRY
14
15
16			class AbstractUNet(Backbone):
17			"""An interface for u-net style backbones."""
18
19			def __init__(
			0 ignored issues – show introduced 2021-02-17 22:43 UTC by Report Bug Copy Issue Report "use_additive_upsampling" differing in parameter documentation Loading history...
20			self,
21			image_size: tuple,
22			num_channel_initial: int,
23			depth: int,
24			extract_levels: List[int],
25			out_kernel_initializer: str,
26			out_activation: str,
27			out_channels: int,
28			name: str = "AbstractUNet",
29			**kwargs,
30			):
31			"""
32			Init.
33
34			Image is encoded gradually, i from level 0 to D,
35			then it is decoded gradually, j from level D to 0.
36			Some of the decoded levels are used for generating extractions.
37
38			So, extract_levels are between [0, D].
39
40			:param image_size: such as (dim1, dim2, dim3)
41			:param num_channel_initial: number of initial channels.
42			:param depth: d = 0 to depth, both side included
43			:param extract_levels: from which depths the output will be built.
44			:param out_kernel_initializer: initializer to use for kernels.
45			:param out_activation: activation to use at end layer.
46			:param out_channels: number of channels for the extractions
47			:param use_additive_upsampling: whether use additive up-sampling.
48			:param name: name of the backbone.
49			:param kwargs: additional arguments.
50			"""
51			super().__init__(
52			image_size=image_size,
53			out_channels=out_channels,
54			num_channel_initial=num_channel_initial,
55			out_kernel_initializer=out_kernel_initializer,
56			out_activation=out_activation,
57			name=name,
58			**kwargs,
59			)
60
61			# save parameters
62			assert max(extract_levels) <= depth
63			self._extract_levels = extract_levels
64			self._depth = depth
65
66			# init layers
67			# all lists start with d = 0
68			self._downsample_convs = None
69			self._downsample_pools = None
70			self._bottom_block = None
71			self._upsample_deconvs = None
72			self._upsample_convs = None
73			self._output_block = None
74
75			def build_layers(
76			self,
77			image_size: tuple,
78			num_channel_initial: int,
79			depth: int,
80			extract_levels: List[int],
81			downsample_kernel_sizes: Union[int, List[int]],
82			upsample_kernel_sizes: Union[int, List[int]],
83			strides: int,
84			padding: str,
85			out_kernel_initializer: str,
86			out_activation: str,
87			out_channels: int,
88			):
89			"""
90			Build layers that will be used in call.
91
92			:param image_size: (dim1, dim2, dim3).
93			:param num_channel_initial: number of initial channels.
94			:param depth: network starts with d = 0, and the bottom has d = depth.
95			:param extract_levels: from which depths the output will be built.
96			:param downsample_kernel_sizes: kernel size for down-sampling
97			:param upsample_kernel_sizes: kernel size for up-sampling
98			:param strides: strides for down-sampling
99			:param padding: padding mode for all conv layers
100			:param out_kernel_initializer: initializer to use for kernels.
101			:param out_activation: activation to use at end layer.
102			:param out_channels: number of channels for the extractions
103			"""
104			# init params
105			min_extract_level = min(extract_levels)
106			num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]
107			if isinstance(downsample_kernel_sizes, int):
108			downsample_kernel_sizes = [downsample_kernel_sizes] * (depth + 1)
109			assert len(downsample_kernel_sizes) == depth + 1
110			if isinstance(upsample_kernel_sizes, int):
111			upsample_kernel_sizes = [upsample_kernel_sizes] * depth
112			assert len(upsample_kernel_sizes) == depth
113
114			# down-sampling
115			self._downsample_convs = []
116			self._downsample_pools = []
117			tensor_shape = image_size
118			tensor_shapes = [tensor_shape]
119			for d in range(depth):
120			downsample_conv = self.build_conv_block(
121			filters=num_channels[d],
122			kernel_size=downsample_kernel_sizes[d],
123			padding=padding,
124			)
125			downsample_pool = self.build_down_sampling_block(
126			kernel_size=strides, strides=strides, padding=padding
127			)
128			tensor_shape = tuple(
129			conv_utils.conv_output_length(
130			input_length=x,
131			filter_size=strides,
132			padding=padding,
133			stride=strides,
134			dilation=1,
135			)
136			for x in tensor_shape
137			)
138			self._downsample_convs.append(downsample_conv)
139			self._downsample_pools.append(downsample_pool)
140			tensor_shapes.append(tensor_shape)
141
142			# bottom layer
143			self._bottom_block = self.build_bottom_block(
144			filters=num_channels[depth],
145			kernel_size=downsample_kernel_sizes[depth],
146			padding=padding,
147			)
148
149			# up-sampling
150			self._upsample_deconvs = []
151			self._upsample_convs = []
152			for d in range(depth - 1, min_extract_level - 1, -1):
153			kernel_size = upsample_kernel_sizes[d]
154			output_padding = layer_util.deconv_output_padding(
155			input_shape=tensor_shapes[d + 1],
156			output_shape=tensor_shapes[d],
157			kernel_size=kernel_size,
158			stride=strides,
159			padding=padding,
160			)
161			upsample_deconv = self.build_up_sampling_block(
162			filters=num_channels[d],
163			output_padding=output_padding,
164			kernel_size=kernel_size,
165			strides=strides,
166			padding=padding,
167			output_shape=tensor_shapes[d],
168			)
169			upsample_conv = self.build_conv_block(
170			filters=num_channels[d], kernel_size=kernel_size, padding=padding
171			)
172			self._upsample_deconvs = [upsample_deconv] + self._upsample_deconvs
173			self._upsample_convs = [upsample_conv] + self._upsample_convs
174			if min_extract_level > 0:
175			# add Nones to make lists have length depth - 1
176			self._upsample_deconvs = [None] * min_extract_level + self._upsample_deconvs
177			self._upsample_convs = [None] * min_extract_level + self._upsample_convs
178
179			# extraction
180			self._output_block = self.build_output_block(
181			image_size=image_size,
182			extract_levels=extract_levels,
183			out_channels=out_channels,
184			out_kernel_initializer=out_kernel_initializer,
185			out_activation=out_activation,
186			)
187
188			@abstractmethod
189			def build_conv_block(
190			self, filters: int, kernel_size: int, padding: str
191			) -> Union[tf.keras.Model, tfkl.Layer]:
192			"""
193			Build a conv block for down-sampling or up-sampling.
194
195			This block do not change the tensor shape (width, height, depth),
196			it only changes the number of channels.
197
198			:param filters: number of channels for output
199			:param kernel_size: arg for conv3d
200			:param padding: arg for conv3d
201			:return: a block consists of one or multiple layers
202			"""
203
204			@abstractmethod
205			def build_down_sampling_block(
206			self, kernel_size: int, padding: str, strides: int
207			) -> Union[tf.keras.Model, tfkl.Layer]:
208			"""
209			Build a block for down-sampling.
210
211			This block changes the tensor shape (width, height, depth),
212			but it does not changes the number of channels.
213
214			:param kernel_size: arg for pool3d
215			:param padding: arg for pool3d
216			:param strides: arg for pool3d
217			:return: a block consists of one or multiple layers
218			"""
219
220			@abstractmethod
221			def build_bottom_block(
222			self, filters: int, kernel_size: int, padding: str
223			) -> Union[tf.keras.Model, tfkl.Layer]:
224			"""
225			Build a block for bottom layer.
226
227			This block do not change the tensor shape (width, height, depth),
228			it only changes the number of channels.
229
230			:param filters: number of channels for output
231			:param kernel_size: arg for conv3d
232			:param padding: arg for conv3d
233			:return: a block consists of one or multiple layers
234			"""
235
236			@abstractmethod
237			def build_up_sampling_block(
238			self,
239			filters: int,
240			output_padding: int,
241			kernel_size: int,
242			padding: str,
243			strides: int,
244			output_shape: tuple,
245			) -> Union[tf.keras.Model, tfkl.Layer]:
246			"""
247			Build a block for up-sampling.
248
249			This block changes the tensor shape (width, height, depth),
250			but it does not changes the number of channels.
251
252			:param filters: number of channels for output
253			:param output_padding: padding for output
254			:param kernel_size: arg for deconv3d
255			:param padding: arg for deconv3d
256			:param strides: arg for deconv3d
257			:param output_shape: shape of the output tensor
258			:return: a block consists of one or multiple layers
259			"""
260
261			@abstractmethod
262			def build_skip_block(self) -> Union[tf.keras.Model, tfkl.Layer]:
263			"""
264			Build a block for combining skipped tensor and up-sampled one.
265
266			This block do not change the tensor shape (width, height, depth),
267			it only changes the number of channels.
268
269			The input to this block is a list of tensors.
270
271			:return: a block consists of one or multiple layers
272			"""
273
274			@abstractmethod
275			def build_output_block(
276			self,
277			image_size: Tuple[int],
278			extract_levels: List[int],
279			out_channels: int,
280			out_kernel_initializer: str,
281			out_activation: str,
282			) -> Union[tf.keras.Model, tfkl.Layer]:
283			"""
284			Build a block for output.
285
286			The input to this block is a list of tensors.
287
288			:param image_size: such as (dim1, dim2, dim3)
289			:param extract_levels: number of extraction levels.
290			:param out_channels: number of channels for the extractions
291			:param out_kernel_initializer: initializer to use for kernels.
292			:param out_activation: activation to use at end layer.
293			:return: a block consists of one or multiple layers
294			"""
295
296			def call(self, inputs: tf.Tensor, training=None, mask=None) -> tf.Tensor:
			0 ignored issues – show introduced 2021-01-26 01:41 UTC by Report Bug Copy Issue Report "mask, training" missing in parameter type documentation Loading history...
297			"""
298			Build LocalNet graph based on built layers.
299
300			:param inputs: image batch, shape = (batch, f_dim1, f_dim2, f_dim3, ch)
301			:param training: None or bool.
302			:param mask: None or tf.Tensor.
303			:return: shape = (batch, f_dim1, f_dim2, f_dim3, out_channels)
304			"""
305
306			# down-sampling
307			skips = []
308			down_sampled = inputs
309			for d in range(self._depth):
310			skip = self._downsample_convs[d](inputs=down_sampled, training=training)
311			down_sampled = self._downsample_pools[d](inputs=skip, training=training)
312			skips.append(skip)
313
314			# bottom
315			up_sampled = self._bottom_block(inputs=down_sampled, training=training)
316
317			# up-sampling
318			outs = [up_sampled]
319			for d in range(self._depth - 1, min(self._extract_levels) - 1, -1):
320			up_sampled = self._upsample_deconvs[d](inputs=up_sampled, training=training)
321			up_sampled = self.build_skip_block()([up_sampled, skips[d]])
322			up_sampled = self._upsample_convs[d](inputs=up_sampled, training=training)
323			outs.append(up_sampled)
324
325			# output
326			output = self._output_block(outs)
327
328			return output
329
330
331			@REGISTRY.register_backbone(name="unet")
332			class UNet(Backbone):
333			"""
334			Class that implements an adapted 3D UNet.
335
336			Reference:
337
338			- O. Ronneberger, P. Fischer, and T. Brox,
339			“U-net: Convolutional networks for biomedical image segmentation,”,
340			Lecture Notes in Computer Science, 2015, vol. 9351, pp. 234–241.
341			https://arxiv.org/abs/1505.04597
342			"""
343
344			def __init__(
345			self,
346			image_size: tuple,
347			out_channels: int,
348			num_channel_initial: int,
349			depth: int,
350			out_kernel_initializer: str,
351			out_activation: str,
352			pooling: bool = True,
353			concat_skip: bool = False,
354			name: str = "Unet",
355			**kwargs,
356			):
357			"""
358			Initialise UNet.
359
360			:param image_size: (dim1, dim2, dim3), dims of input image.
361			:param out_channels: number of channels for the output
362			:param num_channel_initial: number of initial channels
363			:param depth: input is at level 0, bottom is at level depth
364			:param out_kernel_initializer: kernel initializer for the last layer
365			:param out_activation: activation at the last layer
366			:param pooling: for downsampling, use non-parameterized
367			pooling if true, otherwise use conv3d
368			:param concat_skip: when upsampling, concatenate skipped
369			tensor if true, otherwise use addition
370			:param name: name of the backbone.
371			:param kwargs: additional arguments.
372			"""
373			super().__init__(
374			image_size=image_size,
375			out_channels=out_channels,
376			num_channel_initial=num_channel_initial,
377			out_kernel_initializer=out_kernel_initializer,
378			out_activation=out_activation,
379			name=name,
380			**kwargs,
381			)
382
383			# init layer variables
384			num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]
385
386			self._num_channel_initial = num_channel_initial
387			self._depth = depth
388			self._concat_skip = concat_skip
389			self._downsample_convs = []
390			self._downsample_pools = []
391			tensor_shape = image_size
392			self._tensor_shapes = [tensor_shape]
393			for d in range(depth):
394			downsample_conv = tf.keras.Sequential(
395			[
396			layer.Conv3dBlock(
397			filters=num_channels[d], kernel_size=3, padding="same"
398			),
399			layer.ResidualConv3dBlock(
400			filters=num_channels[d], kernel_size=3, padding="same"
401			),
402			]
403			)
404			if pooling:
405			downsample_pool = tfkl.MaxPool3D(pool_size=2, strides=2, padding="same")
406			else:
407			downsample_pool = layer.Conv3dBlock(
408			filters=num_channels[d], kernel_size=3, strides=2, padding="same"
409			)
410			tensor_shape = tuple((x + 1) // 2 for x in tensor_shape)
411			self._downsample_convs.append(downsample_conv)
412			self._downsample_pools.append(downsample_pool)
413			self._tensor_shapes.append(tensor_shape)
414			self._bottom_conv3d = layer.Conv3dBlock(
415			filters=num_channels[depth], kernel_size=3, padding="same"
416			)
417			self._bottom_res3d = layer.ResidualConv3dBlock(
418			filters=num_channels[depth], kernel_size=3, padding="same"
419			)
420			self._upsample_deconvs = []
421			self._upsample_convs = []
422			for d in range(depth):
423			padding = layer_util.deconv_output_padding(
424			input_shape=self._tensor_shapes[d + 1],
425			output_shape=self._tensor_shapes[d],
426			kernel_size=3,
427			stride=2,
428			padding="same",
429			)
430			upsample_deconv = layer.Deconv3dBlock(
431			filters=num_channels[d],
432			output_padding=padding,
433			kernel_size=3,
434			strides=2,
435			padding="same",
436			)
437			upsample_conv = tf.keras.Sequential(
438			[
439			layer.Conv3dBlock(
440			filters=num_channels[d], kernel_size=3, padding="same"
441			),
442			layer.ResidualConv3dBlock(
443			filters=num_channels[d], kernel_size=3, padding="same"
444			),
445			]
446			)
447			self._upsample_deconvs.append(upsample_deconv)
448			self._upsample_convs.append(upsample_conv)
449			self._output_conv3d = tf.keras.Sequential(
450			[
451			tfkl.Conv3D(
452			filters=out_channels,
453			kernel_size=3,
454			strides=1,
455			padding="same",
456			kernel_initializer=out_kernel_initializer,
457			activation=out_activation,
458			),
459			layer.Resize3d(shape=image_size),
460			]
461			)
462
463			def call(self, inputs: tf.Tensor, training=None, mask=None) -> tf.Tensor:
			0 ignored issues – show introduced 2021-01-26 01:41 UTC by Report Bug Copy Issue Report "mask, training" missing in parameter type documentation Loading history...
464			"""
465			Builds graph based on built layers.
466
467			:param inputs: shape = [batch, f_dim1, f_dim2, f_dim3, in_channels]
468			:param training:
469			:param mask:
470			:return: shape = [batch, f_dim1, f_dim2, f_dim3, out_channels]
471			"""
472
473			down_sampled = inputs
474
475			# down sample
476			skips = []
477			for d_var in range(self._depth): # level 0 to D-1
478			skip = self._downsample_convs[d_var](inputs=down_sampled, training=training)
479			down_sampled = self._downsample_pools[d_var](inputs=skip, training=training)
480			skips.append(skip)
481
482			# bottom, level D
483			up_sampled = self._bottom_res3d(
484			inputs=self._bottom_conv3d(inputs=down_sampled, training=training),
485			training=training,
486			)
487
488			# up sample, level D-1 to 0
489			for d_var in range(self._depth - 1, -1, -1):
490			up_sampled = self._upsample_deconvs[d_var](
491			inputs=up_sampled, training=training
492			)
493			if self._concat_skip:
494			up_sampled = tf.concat([up_sampled, skips[d_var]], axis=4)
495			else:
496			up_sampled = up_sampled + skips[d_var]
497			up_sampled = self._upsample_convs[d_var](
498			inputs=up_sampled, training=training
499			)
500
501			# output
502			output = self._output_conv3d(inputs=up_sampled, training=training)
503
504			return output
505

DeepRegNet / DeepReg

Pull Request — main (#656)

AbstractUNet.build_layers() C

Complexity

Size

Duplication

Importance

How to fix Long Method Many Parameters

Long Method

Many Parameters

Duplication Side-by-Side

Filter issues like