deepreg.model.backbone.u_net.UNet.call() - 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 2021-02-01 00:19 UTC

deepreg.model.backbone.u_net.UNet.call() A

↳ Parent: deepreg.model.backbone.u_net

Complexity

Conditions

Size

Total Lines	39
Code Lines	18

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	18
dl	0
loc	39
rs	9.5
c	0
b	0
f	0
cc	4
nop	4

# coding=utf-8



import tensorflow as tf
import tensorflow.keras.layers as tfkl

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


@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,
        control_points: (tuple, None) = None,
        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 control_points: specify the distance between control points (in voxels).
        :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._downsample_convs = []
        self._downsample_pools = []
        self._upsample_blocks = []
        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"
        )
        for d in range(depth):
            padding = layer.deconv_output_padding(
                input_shape=self._tensor_shapes[d + 1],
                output_shape=self._tensor_shapes[d],
                kernel_size=3,
                stride=2,
                padding="same",
            )
            upsample_block = layer.UpSampleResnetBlock(
                filters=num_channels[d], output_padding=padding, concat=concat_skip
            )
            self._upsample_blocks.append(upsample_block)
            self._tensor_shapes.append(tensor_shape)
        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),
            ]
        )

        self.resize = (
            layer.ResizeCPTransform(control_points)
            if control_points is not None
            else False
        )
        self.interpolate = (
            layer.BSplines3DTransform(control_points, image_size)
            if control_points is not None
            else False
        )

    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)
            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_blocks[d_var](
                inputs=[up_sampled, skips[d_var]], training=training
            )

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

        if self.resize:
            output = self.resize(output)
            output = self.interpolate(output)

        return output


1			# coding=utf-8
			0 ignored issues – show introduced 2021-01-26 01:41 UTC by Report Bug Copy Issue Report Missing module docstring Loading history...
2
3
4			import tensorflow as tf
5			import tensorflow.keras.layers as tfkl
6
7			from deepreg.model import layer
8			from deepreg.model.backbone.interface import Backbone
9			from deepreg.registry import REGISTRY
10
11
12			@REGISTRY.register_backbone(name="unet")
13			class UNet(Backbone):
14			"""
15			Class that implements an adapted 3D UNet.
16
17			Reference:
18
19			- O. Ronneberger, P. Fischer, and T. Brox,
20			“U-net: Convolutional networks for biomedical image segmentation,”,
21			Lecture Notes in Computer Science, 2015, vol. 9351, pp. 234–241.
22			https://arxiv.org/abs/1505.04597
23			"""
24
25			def __init__(
26			self,
27			image_size: tuple,
28			out_channels: int,
29			num_channel_initial: int,
30			depth: int,
31			out_kernel_initializer: str,
32			out_activation: str,
33			pooling: bool = True,
34			concat_skip: bool = False,
35			control_points: (tuple, None) = None,
36			name: str = "Unet",
37			**kwargs,
38			):
39			"""
40			Initialise UNet.
41
42			:param image_size: (dim1, dim2, dim3), dims of input image.
43			:param out_channels: number of channels for the output
44			:param num_channel_initial: number of initial channels
45			:param depth: input is at level 0, bottom is at level depth
46			:param out_kernel_initializer: kernel initializer for the last layer
47			:param out_activation: activation at the last layer
48			:param pooling: for downsampling, use non-parameterized
49			pooling if true, otherwise use conv3d
50			:param concat_skip: when upsampling, concatenate skipped
51			tensor if true, otherwise use addition
52			:param control_points: specify the distance between control points (in voxels).
53			:param name: name of the backbone.
54			:param kwargs: additional arguments.
55			"""
56			super().__init__(
57			image_size=image_size,
58			out_channels=out_channels,
59			num_channel_initial=num_channel_initial,
60			out_kernel_initializer=out_kernel_initializer,
61			out_activation=out_activation,
62			name=name,
63			**kwargs,
64			)
65
66			# init layer variables
67			num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]
68
69			self._num_channel_initial = num_channel_initial
70			self._depth = depth
71			self._downsample_convs = []
72			self._downsample_pools = []
73			self._upsample_blocks = []
74			tensor_shape = image_size
75			self._tensor_shapes = [tensor_shape]
76			for d in range(depth):
77			downsample_conv = tf.keras.Sequential(
78			[
79			layer.Conv3dBlock(
80			filters=num_channels[d], kernel_size=3, padding="same"
81			),
82			layer.ResidualConv3dBlock(
83			filters=num_channels[d], kernel_size=3, padding="same"
84			),
85			]
86			)
87			if pooling:
88			downsample_pool = tfkl.MaxPool3D(pool_size=2, strides=2, padding="same")
89			else:
90			downsample_pool = layer.Conv3dBlock(
91			filters=num_channels[d], kernel_size=3, strides=2, padding="same"
92			)
93			tensor_shape = tuple((x + 1) // 2 for x in tensor_shape)
94			self._downsample_convs.append(downsample_conv)
95			self._downsample_pools.append(downsample_pool)
96			self._tensor_shapes.append(tensor_shape)
97			self._bottom_conv3d = layer.Conv3dBlock(
98			filters=num_channels[depth], kernel_size=3, padding="same"
99			)
100			self._bottom_res3d = layer.ResidualConv3dBlock(
101			filters=num_channels[depth], kernel_size=3, padding="same"
102			)
103			for d in range(depth):
104			padding = layer.deconv_output_padding(
105			input_shape=self._tensor_shapes[d + 1],
106			output_shape=self._tensor_shapes[d],
107			kernel_size=3,
108			stride=2,
109			padding="same",
110			)
111			upsample_block = layer.UpSampleResnetBlock(
112			filters=num_channels[d], output_padding=padding, concat=concat_skip
113			)
114			self._upsample_blocks.append(upsample_block)
115			self._tensor_shapes.append(tensor_shape)
116			self._output_conv3d = tf.keras.Sequential(
117			[
118			tfkl.Conv3D(
119			filters=out_channels,
120			kernel_size=3,
121			strides=1,
122			padding="same",
123			kernel_initializer=out_kernel_initializer,
124			activation=out_activation,
125			),
126			layer.Resize3d(shape=image_size),
127			]
128			)
129
130			self.resize = (
131			layer.ResizeCPTransform(control_points)
132			if control_points is not None
133			else False
134			)
135			self.interpolate = (
136			layer.BSplines3DTransform(control_points, image_size)
137			if control_points is not None
138			else False
139			)
140
141			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...
142			"""
143			Builds graph based on built layers.
144
145			:param inputs: shape = [batch, f_dim1, f_dim2, f_dim3, in_channels]
146			:param training:
147			:param mask:
148			:return: shape = [batch, f_dim1, f_dim2, f_dim3, out_channels]
149			"""
150
151			down_sampled = inputs
152
153			# down sample
154			skips = []
155			for d_var in range(self._depth): # level 0 to D-1
156			skip = self._downsample_convs[d_var](inputs=down_sampled, training=training)
157			down_sampled = self._downsample_pools[d_var](inputs=skip)
158			skips.append(skip)
159
160			# bottom, level D
161			up_sampled = self._bottom_res3d(
162			inputs=self._bottom_conv3d(inputs=down_sampled, training=training),
163			training=training,
164			)
165
166			# up sample, level D-1 to 0
167			for d_var in range(self._depth - 1, -1, -1):
168			up_sampled = self._upsample_blocks[d_var](
169			inputs=[up_sampled, skips[d_var]], training=training
170			)
171
172			# output
173			output = self._output_conv3d(inputs=up_sampled)
174
175			if self.resize:
176			output = self.resize(output)
177			output = self.interpolate(output)
178
179			return output
180

DeepRegNet / DeepReg

Pull Request — main (#656)

deepreg.model.backbone.u_net.UNet.call() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like