deepreg.model.backbone.u_net

Complexity

Total Complexity

8

Size/Duplication

Total Lines	187
Duplicated Lines	0 %

Importance

Changes

0

2 Methods

Rating	Name	Duplication	Size	Complexity
B	UNet.__init__()	0	116	4
A	UNet.call()	0	42	4

# coding=utf-8

Missing module docstring

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:

"mask, training" missing in parameter type documentation

        """
        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