deepreg.model.backbone.u_net.UNet.__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 2021-02-02 23:26 UTC

deepreg.model.backbone.u_net.UNet.init() B

↳ Parent: deepreg.model.backbone.u_net

Complexity

Conditions

Size

Total Lines	116
Code Lines	80

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	80
dl	0
loc	116
rs	7.6545
c	0
b	0
f	0
cc	4
nop	11

How to fix Long Method Many Parameters

# coding=utf-8



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

import deepreg.model.layer_util
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,
        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 = deepreg.model.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 2021-01-26 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			import deepreg.model.layer_util
8			from deepreg.model import layer
9			from deepreg.model.backbone.interface import Backbone
10			from deepreg.registry import REGISTRY
11
12
13			@REGISTRY.register_backbone(name="unet")
14			class UNet(Backbone):
15			"""
16			Class that implements an adapted 3D UNet.
17
18			Reference:
19
20			- O. Ronneberger, P. Fischer, and T. Brox,
21			“U-net: Convolutional networks for biomedical image segmentation,”,
22			Lecture Notes in Computer Science, 2015, vol. 9351, pp. 234–241.
23			https://arxiv.org/abs/1505.04597
24			"""
25
26			def __init__(
27			self,
28			image_size: tuple,
29			out_channels: int,
30			num_channel_initial: int,
31			depth: int,
32			out_kernel_initializer: str,
33			out_activation: str,
34			pooling: bool = True,
35			concat_skip: bool = False,
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 name: name of the backbone.
53			:param kwargs: additional arguments.
54			"""
55			super().__init__(
56			image_size=image_size,
57			out_channels=out_channels,
58			num_channel_initial=num_channel_initial,
59			out_kernel_initializer=out_kernel_initializer,
60			out_activation=out_activation,
61			name=name,
62			**kwargs,
63			)
64
65			# init layer variables
66			num_channels = [num_channel_initial * (2 ** d) for d in range(depth + 1)]
67
68			self._num_channel_initial = num_channel_initial
69			self._depth = depth
70			self._concat_skip = concat_skip
71			self._downsample_convs = []
72			self._downsample_pools = []
73			tensor_shape = image_size
74			self._tensor_shapes = [tensor_shape]
75			for d in range(depth):
76			downsample_conv = tf.keras.Sequential(
77			[
78			layer.Conv3dBlock(
79			filters=num_channels[d], kernel_size=3, padding="same"
80			),
81			layer.ResidualConv3dBlock(
82			filters=num_channels[d], kernel_size=3, padding="same"
83			),
84			]
85			)
86			if pooling:
87			downsample_pool = tfkl.MaxPool3D(pool_size=2, strides=2, padding="same")
88			else:
89			downsample_pool = layer.Conv3dBlock(
90			filters=num_channels[d], kernel_size=3, strides=2, padding="same"
91			)
92			tensor_shape = tuple((x + 1) // 2 for x in tensor_shape)
93			self._downsample_convs.append(downsample_conv)
94			self._downsample_pools.append(downsample_pool)
95			self._tensor_shapes.append(tensor_shape)
96			self._bottom_conv3d = layer.Conv3dBlock(
97			filters=num_channels[depth], kernel_size=3, padding="same"
98			)
99			self._bottom_res3d = layer.ResidualConv3dBlock(
100			filters=num_channels[depth], kernel_size=3, padding="same"
101			)
102			self._upsample_deconvs = []
103			self._upsample_convs = []
104			for d in range(depth):
105			padding = deepreg.model.layer_util.deconv_output_padding(
106			input_shape=self._tensor_shapes[d + 1],
107			output_shape=self._tensor_shapes[d],
108			kernel_size=3,
109			stride=2,
110			padding="same",
111			)
112			upsample_deconv = layer.Deconv3dBlock(
113			filters=num_channels[d],
114			output_padding=padding,
115			kernel_size=3,
116			strides=2,
117			padding="same",
118			)
119			upsample_conv = tf.keras.Sequential(
120			[
121			layer.Conv3dBlock(
122			filters=num_channels[d], kernel_size=3, padding="same"
123			),
124			layer.ResidualConv3dBlock(
125			filters=num_channels[d], kernel_size=3, padding="same"
126			),
127			]
128			)
129			self._upsample_deconvs.append(upsample_deconv)
130			self._upsample_convs.append(upsample_conv)
131			self._output_conv3d = tf.keras.Sequential(
132			[
133			tfkl.Conv3D(
134			filters=out_channels,
135			kernel_size=3,
136			strides=1,
137			padding="same",
138			kernel_initializer=out_kernel_initializer,
139			activation=out_activation,
140			),
141			layer.Resize3d(shape=image_size),
142			]
143			)
144
145			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...
146			"""
147			Builds graph based on built layers.
148
149			:param inputs: shape = [batch, f_dim1, f_dim2, f_dim3, in_channels]
150			:param training:
151			:param mask:
152			:return: shape = [batch, f_dim1, f_dim2, f_dim3, out_channels]
153			"""
154
155			down_sampled = inputs
156
157			# down sample
158			skips = []
159			for d_var in range(self._depth): # level 0 to D-1
160			skip = self._downsample_convs[d_var](inputs=down_sampled, training=training)
161			down_sampled = self._downsample_pools[d_var](inputs=skip, training=training)
162			skips.append(skip)
163
164			# bottom, level D
165			up_sampled = self._bottom_res3d(
166			inputs=self._bottom_conv3d(inputs=down_sampled, training=training),
167			training=training,
168			)
169
170			# up sample, level D-1 to 0
171			for d_var in range(self._depth - 1, -1, -1):
172			up_sampled = self._upsample_deconvs[d_var](
173			inputs=up_sampled, training=training
174			)
175			if self._concat_skip:
176			up_sampled = tf.concat([up_sampled, skips[d_var]], axis=4)
177			else:
178			up_sampled = up_sampled + skips[d_var]
179			up_sampled = self._upsample_convs[d_var](
180			inputs=up_sampled, training=training
181			)
182
183			# output
184			output = self._output_conv3d(inputs=up_sampled, training=training)
185
186			return output
187

DeepRegNet / DeepReg

Pull Request — main (#656)

deepreg.model.backbone.u_net.UNet.__init__() B

Complexity

Size

Duplication

Importance

How to fix Long Method Many Parameters

Long Method

Many Parameters

Duplication Side-by-Side

Filter issues like

deepreg.model.backbone.u_net.UNet.init() B