deepreg.model.network.RegistrationModel.log_tensor_stats() - Code Metrics - Inspection of "708 log more metrics in tensorboard" - DeepRegNet/DeepReg - Measure and Improve Code Quality continuously with Scrutinizer

Passed

Pull Request — main (#713)

by Yunguan

created 2021-03-23 07:56 UTC

RegistrationModel.log_tensor_stats() A

↳ Parent: deepreg.model.network

Complexity

Conditions

Size

Total Lines	22
Code Lines	14

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	14
dl	0
loc	22
rs	9.7
c	0
b	0
f	0
cc	1
nop	3

import logging
import os
from abc import abstractmethod
from copy import deepcopy
from typing import Dict, Optional, Tuple

import tensorflow as tf

from deepreg.loss.label import compute_centroid_distance
from deepreg.model import layer
from deepreg.model.backbone import GlobalNet
from deepreg.registry import REGISTRY


def dict_without(d: dict, key) -> dict:
    """
    Return a copy of the given dict without a certain key.

    :param d: dict to be copied.
    :param key: key to be removed.
    :return: the copy without a key
    """
    copied = deepcopy(d)
    copied.pop(key)
    return copied


class RegistrationModel(tf.keras.Model):
    """Interface for registration model."""

    def __init__(
        self,
        moving_image_size: tuple,
        fixed_image_size: tuple,
        index_size: int,
        labeled: bool,
        batch_size: int,
        config: dict,
        num_devices: int = 1,
        name: str = "RegistrationModel",
    ):
        """
        Init.

        :param moving_image_size: (m_dim1, m_dim2, m_dim3)
        :param fixed_image_size: (f_dim1, f_dim2, f_dim3)
        :param index_size: number of indices for identify each sample
        :param labeled: if the data is labeled
        :param batch_size: size of mini-batch
        :param config: config for method, backbone, and loss.
        :param num_devices: number of GPU used,
            global_batch_size = batch_size*num_devices
        :param name: name of the model
        """
        super().__init__(name=name)
        self.moving_image_size = moving_image_size
        self.fixed_image_size = fixed_image_size
        self.index_size = index_size
        self.labeled = labeled
        self.batch_size = batch_size
        self.config = config
        self.num_devices = num_devices
        self.global_batch_size = num_devices * batch_size

        self._inputs = None  # save inputs of self._model as dict
        self._outputs = None  # save outputs of self._model as dict

        self._model: tf.keras.Model = self.build_model()
        self.build_loss()

    def get_config(self) -> dict:
        """Return the config dictionary for recreating this class."""
        return dict(
            moving_image_size=self.moving_image_size,
            fixed_image_size=self.fixed_image_size,
            index_size=self.index_size,
            labeled=self.labeled,
            batch_size=self.batch_size,
            config=self.config,
            num_devices=self.num_devices,
            name=self.name,
        )

    @abstractmethod
    def build_model(self):
        """Build the model to be saved as self._model."""

    def build_inputs(self) -> Dict[str, tf.keras.layers.Input]:
        """
        Build input tensors.

        :return: dict of inputs.
        """
        # (batch, m_dim1, m_dim2, m_dim3, 1)
        moving_image = tf.keras.Input(
            shape=self.moving_image_size,
            batch_size=self.batch_size,
            name="moving_image",
        )
        # (batch, f_dim1, f_dim2, f_dim3, 1)
        fixed_image = tf.keras.Input(
            shape=self.fixed_image_size,
            batch_size=self.batch_size,
            name="fixed_image",
        )
        # (batch, index_size)
        indices = tf.keras.Input(
            shape=(self.index_size,),
            batch_size=self.batch_size,
            name="indices",
        )

        if not self.labeled:
            return dict(
                moving_image=moving_image, fixed_image=fixed_image, indices=indices
            )

        # (batch, m_dim1, m_dim2, m_dim3, 1)
        moving_label = tf.keras.Input(
            shape=self.moving_image_size,
            batch_size=self.batch_size,
            name="moving_label",
        )
        # (batch, m_dim1, m_dim2, m_dim3, 1)
        fixed_label = tf.keras.Input(
            shape=self.fixed_image_size,
            batch_size=self.batch_size,
            name="fixed_label",
        )
        return dict(
            moving_image=moving_image,
            fixed_image=fixed_image,
            moving_label=moving_label,
            fixed_label=fixed_label,
            indices=indices,
        )

    def concat_images(
        self,
        moving_image: tf.Tensor,
        fixed_image: tf.Tensor,
        moving_label: Optional[tf.Tensor] = None,
    ) -> tf.Tensor:
        """
        Adjust image shape and concatenate them together.

        :param moving_image: registration source
        :param fixed_image: registration target
        :param moving_label: optional, only used for conditional model.
        :return:
        """
        images = []

        resize_layer = layer.Resize3d(shape=self.fixed_image_size)

        # (batch, m_dim1, m_dim2, m_dim3, 1)
        moving_image = tf.expand_dims(moving_image, axis=4)
        moving_image = resize_layer(moving_image)
        images.append(moving_image)

        # (batch, m_dim1, m_dim2, m_dim3, 1)
        fixed_image = tf.expand_dims(fixed_image, axis=4)
        images.append(fixed_image)

        # (batch, m_dim1, m_dim2, m_dim3, 1)
        if moving_label is not None:
            moving_label = tf.expand_dims(moving_label, axis=4)
            moving_label = resize_layer(moving_label)
            images.append(moving_label)

        # (batch, f_dim1, f_dim2, f_dim3, 2 or 3)
        images = tf.concat(images, axis=4)
        return images

    def _build_loss(self, name: str, inputs_dict: dict):
        """
        Build and add one weighted loss together with the metrics.

        :param name: name of loss, image / label / regularization.
        :param inputs_dict: inputs for loss function
        """

        if name not in self.config["loss"]:
            # loss config is not defined
            logging.warning(
                f"The configuration for loss {name} is not defined. "
                f"Therefore it is not used."
            )
            return

        loss_configs = self.config["loss"][name]
        if not isinstance(loss_configs, list):
            loss_configs = [loss_configs]

        for loss_config in loss_configs:

            if "weight" not in loss_config:
                # default loss weight 1
                logging.warning(
                    f"The weight for loss {name} is not defined."
                    f"Default weight = 1.0 is used."
                )
                loss_config["weight"] = 1.0

            # build loss
            weight = loss_config["weight"]

            if weight == 0:
                logging.warning(
                    f"The weight for loss {name} is zero." f"Loss is not used."
                )
                return

            loss_layer: tf.keras.layers.Layer = REGISTRY.build_loss(
                config=dict_without(d=loss_config, key="weight")
            )
            loss_value = loss_layer(**inputs_dict) / self.global_batch_size
            weighted_loss = loss_value * weight

            # add loss
            self._model.add_loss(weighted_loss)

            # add metric
            self._model.add_metric(
                loss_value, name=f"loss/{name}_{loss_layer.name}", aggregation="mean"
            )
            self._model.add_metric(
                weighted_loss,
                name=f"loss/{name}_{loss_layer.name}_weighted",
                aggregation="mean",
            )

    @abstractmethod
    def build_loss(self):
        """Build losses according to configs."""

    def call(
        self, inputs: Dict[str, tf.Tensor], training=None, mask=None
    ) -> Dict[str, tf.Tensor]:
        """
        Call the self._model.

        :param inputs: a dict of tensors.
        :param training: training or not.
        :param mask: maks for inputs.
        :return:
        """
        return self._model(inputs, training=training, mask=mask)  # pragma: no cover

    @abstractmethod
    def postprocess(
        self,
        inputs: Dict[str, tf.Tensor],
        outputs: Dict[str, tf.Tensor],
    ) -> Tuple[tf.Tensor, Dict]:
        """
        Return a dict used for saving inputs and outputs.

        :param inputs: dict of model inputs
        :param outputs: dict of model outputs
        :return: tuple, indices and a dict.
            In the dict, each value is (tensor, normalize, on_label), where
            - normalize = True if the tensor need to be normalized to [0, 1]
            - on_label = True if the tensor depends on label
        """

    def plot_model(self, output_dir: str):
        """
        Save model structure in png.

        :param output_dir: path to the output dir.
        """
        logging.info(self._model.summary())
        try:
            tf.keras.utils.plot_model(
                self._model,
                to_file=os.path.join(output_dir, f"{self.name}.png"),
                dpi=96,
                show_shapes=True,
                show_layer_names=True,
                expand_nested=False,
            )
        except ImportError as err:  # pragma: no cover
            logging.error(
                "Failed to plot model structure."
                "Please check if graphviz is installed.\n"
                "Error message is:"
                f"{err}"
            )

    def log_tensor_stats(self, tensor: tf.Tensor, name: str):
        """
        Log statistics of a given tensor.

        :param tensor: tensor to monitor.
        :param name: name of the tensor.
        """
        flatten = tf.reshape(tensor, shape=(self.batch_size, -1))
        self._model.add_metric(
            tf.reduce_mean(flatten, axis=1),
            name=f"metric/{name}_mean",
            aggregation="mean",
        )
        self._model.add_metric(
            tf.reduce_min(flatten, axis=1),
            name=f"metric/{name}_min",
            aggregation="min",
        )
        self._model.add_metric(
            tf.reduce_max(flatten, axis=1),
            name=f"metric/{name}_max",
            aggregation="max",
        )


@REGISTRY.register_model(name="ddf")
class DDFModel(RegistrationModel):
    """
    A registration model predicts DDF.

    When using global net as backbone,
    the model predicts an affine transformation parameters,
    and a DDF is calculated based on that.
    """

    name = "DDFModel"

    def _resize_interpolate(self, field, control_points):
        resize = layer.ResizeCPTransform(control_points)
        field = resize(field)

        interpolate = layer.BSplines3DTransform(control_points, self.fixed_image_size)
        field = interpolate(field)

        return field

    def build_model(self):
        """Build the model to be saved as self._model."""
        # build inputs
        self._inputs = self.build_inputs()
        moving_image = self._inputs["moving_image"]
        fixed_image = self._inputs["fixed_image"]

        # build ddf
        control_points = self.config["backbone"].pop("control_points", False)
        backbone_inputs = self.concat_images(moving_image, fixed_image)
        backbone = REGISTRY.build_backbone(
            config=self.config["backbone"],
            default_args=dict(
                image_size=self.fixed_image_size,
                out_channels=3,
                out_kernel_initializer="zeros",
                out_activation=None,
            ),
        )

        if isinstance(backbone, GlobalNet):
            # (f_dim1, f_dim2, f_dim3, 3), (4, 3)
            ddf, theta = backbone(inputs=backbone_inputs)
            self._outputs = dict(ddf=ddf, theta=theta)
        else:
            # (f_dim1, f_dim2, f_dim3, 3)
            ddf = backbone(inputs=backbone_inputs)
            ddf = (
                self._resize_interpolate(ddf, control_points) if control_points else ddf
            )
            self._outputs = dict(ddf=ddf)

        # build outputs
        self._warping = layer.Warping(fixed_image_size=self.fixed_image_size)
        # (f_dim1, f_dim2, f_dim3, 3)
        pred_fixed_image = self._warping(inputs=[ddf, moving_image])
        self._outputs["pred_fixed_image"] = pred_fixed_image

        if not self.labeled:
            return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)

        # (f_dim1, f_dim2, f_dim3, 3)
        moving_label = self._inputs["moving_label"]
        pred_fixed_label = self._warping(inputs=[ddf, moving_label])

        self._outputs["pred_fixed_label"] = pred_fixed_label
        return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)

    def build_loss(self):
        """Build losses according to configs."""
        fixed_image = self._inputs["fixed_image"]
        moving_image = self._inputs["moving_image"]
        ddf = self._outputs["ddf"]
        pred_fixed_image = self._outputs["pred_fixed_image"]

        # inputs
        self.log_tensor_stats(tensor=moving_image, name="moving_image")
        self.log_tensor_stats(tensor=fixed_image, name="fixed_image")

        # ddf
        self._build_loss(name="regularization", inputs_dict=dict(inputs=ddf))
        self.log_tensor_stats(tensor=ddf, name="ddf")

        # image
        self._build_loss(
            name="image", inputs_dict=dict(y_true=fixed_image, y_pred=pred_fixed_image)
        )

        # label
        if self.labeled:
            fixed_label = self._inputs["fixed_label"]
            moving_label = self._inputs["moving_label"]
            pred_fixed_label = self._outputs["pred_fixed_label"]
            self._build_loss(
                name="label",
                inputs_dict=dict(y_true=fixed_label, y_pred=pred_fixed_label),
            )
            tre = compute_centroid_distance(
                y_true=fixed_label, y_pred=pred_fixed_label, grid=self._warping.grid_ref
            )
            self._model.add_metric(tre, name="metric/TRE", aggregation="mean")
            self.log_tensor_stats(tensor=moving_label, name="moving_label")
            self.log_tensor_stats(tensor=fixed_label, name="fixed_label")

    def postprocess(
        self,
        inputs: Dict[str, tf.Tensor],
        outputs: Dict[str, tf.Tensor],
    ) -> Tuple[tf.Tensor, Dict]:
        """
        Return a dict used for saving inputs and outputs.

        :param inputs: dict of model inputs
        :param outputs: dict of model outputs
        :return: tuple, indices and a dict.
            In the dict, each value is (tensor, normalize, on_label), where
            - normalize = True if the tensor need to be normalized to [0, 1]
            - on_label = True if the tensor depends on label
        """
        indices = inputs["indices"]
        processed = dict(
            moving_image=(inputs["moving_image"], True, False),
            fixed_image=(inputs["fixed_image"], True, False),
            ddf=(outputs["ddf"], True, False),
            pred_fixed_image=(outputs["pred_fixed_image"], True, False),
        )

        # save theta for affine model
        if "theta" in outputs:
            processed["theta"] = (outputs["theta"], None, None)  # type: ignore

        if not self.labeled:
            return indices, processed

        processed = {
            **dict(
                moving_label=(inputs["moving_label"], False, True),
                fixed_label=(inputs["fixed_label"], False, True),
                pred_fixed_label=(outputs["pred_fixed_label"], False, True),
            ),
            **processed,
        }

        return indices, processed


@REGISTRY.register_model(name="dvf")
class DVFModel(DDFModel):
    """
    A registration model predicts DVF.

    DDF is calculated based on DVF.
    """

    name = "DVFModel"

    def build_model(self):
        """Build the model to be saved as self._model."""
        # build inputs
        self._inputs = self.build_inputs()
        moving_image = self._inputs["moving_image"]
        fixed_image = self._inputs["fixed_image"]
        control_points = self.config["backbone"].pop("control_points", False)

        # build ddf
        backbone_inputs = self.concat_images(moving_image, fixed_image)
        backbone = REGISTRY.build_backbone(
            config=self.config["backbone"],
            default_args=dict(
                image_size=self.fixed_image_size,
                out_channels=3,
                out_kernel_initializer="zeros",
                out_activation=None,
            ),
        )
        dvf = backbone(inputs=backbone_inputs)
        dvf = self._resize_interpolate(dvf, control_points) if control_points else dvf
        ddf = layer.IntDVF(fixed_image_size=self.fixed_image_size)(dvf)

        # build outputs
        self._warping = layer.Warping(fixed_image_size=self.fixed_image_size)
        # (f_dim1, f_dim2, f_dim3, 3)
        pred_fixed_image = self._warping(inputs=[ddf, moving_image])

        self._outputs = dict(dvf=dvf, ddf=ddf, pred_fixed_image=pred_fixed_image)

        if not self.labeled:
            return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)

        # (f_dim1, f_dim2, f_dim3, 3)
        moving_label = self._inputs["moving_label"]
        pred_fixed_label = self._warping(inputs=[ddf, moving_label])

        self._outputs["pred_fixed_label"] = pred_fixed_label
        return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)

    def postprocess(
        self,
        inputs: Dict[str, tf.Tensor],
        outputs: Dict[str, tf.Tensor],
    ) -> Tuple[tf.Tensor, Dict]:
        """
        Return a dict used for saving inputs and outputs.

        :param inputs: dict of model inputs
        :param outputs: dict of model outputs
        :return: tuple, indices and a dict.
            In the dict, each value is (tensor, normalize, on_label), where
            - normalize = True if the tensor need to be normalized to [0, 1]
            - on_label = True if the tensor depends on label
        """
        indices, processed = super().postprocess(inputs=inputs, outputs=outputs)
        processed["dvf"] = (outputs["dvf"], True, False)
        return indices, processed


@REGISTRY.register_model(name="conditional")
class ConditionalModel(RegistrationModel):
    """
    A registration model predicts fixed image label without DDF or DVF.
    """

    name = "ConditionalModel"

    def build_model(self):
        """Build the model to be saved as self._model."""
        assert self.labeled

        # build inputs
        self._inputs = self.build_inputs()
        moving_image = self._inputs["moving_image"]
        fixed_image = self._inputs["fixed_image"]
        moving_label = self._inputs["moving_label"]

        # build ddf
        backbone_inputs = self.concat_images(moving_image, fixed_image, moving_label)
        backbone = REGISTRY.build_backbone(
            config=self.config["backbone"],
            default_args=dict(
                image_size=self.fixed_image_size,
                out_channels=1,
                out_kernel_initializer="glorot_uniform",
                out_activation="sigmoid",
            ),
        )
        # (batch, f_dim1, f_dim2, f_dim3)
        pred_fixed_label = backbone(inputs=backbone_inputs)
        pred_fixed_label = tf.squeeze(pred_fixed_label, axis=4)

        self._outputs = dict(pred_fixed_label=pred_fixed_label)
        return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)

    def build_loss(self):
        """Build losses according to configs."""
        fixed_label = self._inputs["fixed_label"]
        pred_fixed_label = self._outputs["pred_fixed_label"]

        self._build_loss(
            name="label",
            inputs_dict=dict(y_true=fixed_label, y_pred=pred_fixed_label),
        )

    def postprocess(
        self,
        inputs: Dict[str, tf.Tensor],
        outputs: Dict[str, tf.Tensor],
    ) -> Tuple[tf.Tensor, Dict]:
        """
        Return a dict used for saving inputs and outputs.

        :param inputs: dict of model inputs
        :param outputs: dict of model outputs
        :return: tuple, indices and a dict.
            In the dict, each value is (tensor, normalize, on_label), where
            - normalize = True if the tensor need to be normalized to [0, 1]
            - on_label = True if the tensor depends on label
        """
        indices = inputs["indices"]
        processed = dict(
            moving_image=(inputs["moving_image"], True, False),
            fixed_image=(inputs["fixed_image"], True, False),
            pred_fixed_label=(outputs["pred_fixed_label"], True, True),
            moving_label=(inputs["moving_label"], False, True),
            fixed_label=(inputs["fixed_label"], False, True),
        )

        return indices, processed


1			import logging
2			import os
3			from abc import abstractmethod
4			from copy import deepcopy
5			from typing import Dict, Optional, Tuple
6
7			import tensorflow as tf
8
9			from deepreg.loss.label import compute_centroid_distance
10			from deepreg.model import layer
11			from deepreg.model.backbone import GlobalNet
12			from deepreg.registry import REGISTRY
13
14
15			def dict_without(d: dict, key) -> dict:
16			"""
17			Return a copy of the given dict without a certain key.
18
19			:param d: dict to be copied.
20			:param key: key to be removed.
21			:return: the copy without a key
22			"""
23			copied = deepcopy(d)
24			copied.pop(key)
25			return copied
26
27
28			class RegistrationModel(tf.keras.Model):
29			"""Interface for registration model."""
30
31			def __init__(
32			self,
33			moving_image_size: tuple,
34			fixed_image_size: tuple,
35			index_size: int,
36			labeled: bool,
37			batch_size: int,
38			config: dict,
39			num_devices: int = 1,
40			name: str = "RegistrationModel",
41			):
42			"""
43			Init.
44
45			:param moving_image_size: (m_dim1, m_dim2, m_dim3)
46			:param fixed_image_size: (f_dim1, f_dim2, f_dim3)
47			:param index_size: number of indices for identify each sample
48			:param labeled: if the data is labeled
49			:param batch_size: size of mini-batch
50			:param config: config for method, backbone, and loss.
51			:param num_devices: number of GPU used,
52			global_batch_size = batch_size*num_devices
53			:param name: name of the model
54			"""
55			super().__init__(name=name)
56			self.moving_image_size = moving_image_size
57			self.fixed_image_size = fixed_image_size
58			self.index_size = index_size
59			self.labeled = labeled
60			self.batch_size = batch_size
61			self.config = config
62			self.num_devices = num_devices
63			self.global_batch_size = num_devices * batch_size
64
65			self._inputs = None # save inputs of self._model as dict
66			self._outputs = None # save outputs of self._model as dict
67
68			self._model: tf.keras.Model = self.build_model()
69			self.build_loss()
70
71			def get_config(self) -> dict:
72			"""Return the config dictionary for recreating this class."""
73			return dict(
74			moving_image_size=self.moving_image_size,
75			fixed_image_size=self.fixed_image_size,
76			index_size=self.index_size,
77			labeled=self.labeled,
78			batch_size=self.batch_size,
79			config=self.config,
80			num_devices=self.num_devices,
81			name=self.name,
82			)
83
84			@abstractmethod
85			def build_model(self):
86			"""Build the model to be saved as self._model."""
87
88			def build_inputs(self) -> Dict[str, tf.keras.layers.Input]:
89			"""
90			Build input tensors.
91
92			:return: dict of inputs.
93			"""
94			# (batch, m_dim1, m_dim2, m_dim3, 1)
95			moving_image = tf.keras.Input(
96			shape=self.moving_image_size,
97			batch_size=self.batch_size,
98			name="moving_image",
99			)
100			# (batch, f_dim1, f_dim2, f_dim3, 1)
101			fixed_image = tf.keras.Input(
102			shape=self.fixed_image_size,
103			batch_size=self.batch_size,
104			name="fixed_image",
105			)
106			# (batch, index_size)
107			indices = tf.keras.Input(
108			shape=(self.index_size,),
109			batch_size=self.batch_size,
110			name="indices",
111			)
112
113			if not self.labeled:
114			return dict(
115			moving_image=moving_image, fixed_image=fixed_image, indices=indices
116			)
117
118			# (batch, m_dim1, m_dim2, m_dim3, 1)
119			moving_label = tf.keras.Input(
120			shape=self.moving_image_size,
121			batch_size=self.batch_size,
122			name="moving_label",
123			)
124			# (batch, m_dim1, m_dim2, m_dim3, 1)
125			fixed_label = tf.keras.Input(
126			shape=self.fixed_image_size,
127			batch_size=self.batch_size,
128			name="fixed_label",
129			)
130			return dict(
131			moving_image=moving_image,
132			fixed_image=fixed_image,
133			moving_label=moving_label,
134			fixed_label=fixed_label,
135			indices=indices,
136			)
137
138			def concat_images(
139			self,
140			moving_image: tf.Tensor,
141			fixed_image: tf.Tensor,
142			moving_label: Optional[tf.Tensor] = None,
143			) -> tf.Tensor:
144			"""
145			Adjust image shape and concatenate them together.
146
147			:param moving_image: registration source
148			:param fixed_image: registration target
149			:param moving_label: optional, only used for conditional model.
150			:return:
151			"""
152			images = []
153
154			resize_layer = layer.Resize3d(shape=self.fixed_image_size)
155
156			# (batch, m_dim1, m_dim2, m_dim3, 1)
157			moving_image = tf.expand_dims(moving_image, axis=4)
158			moving_image = resize_layer(moving_image)
159			images.append(moving_image)
160
161			# (batch, m_dim1, m_dim2, m_dim3, 1)
162			fixed_image = tf.expand_dims(fixed_image, axis=4)
163			images.append(fixed_image)
164
165			# (batch, m_dim1, m_dim2, m_dim3, 1)
166			if moving_label is not None:
167			moving_label = tf.expand_dims(moving_label, axis=4)
168			moving_label = resize_layer(moving_label)
169			images.append(moving_label)
170
171			# (batch, f_dim1, f_dim2, f_dim3, 2 or 3)
172			images = tf.concat(images, axis=4)
173			return images
174
175			def _build_loss(self, name: str, inputs_dict: dict):
176			"""
177			Build and add one weighted loss together with the metrics.
178
179			:param name: name of loss, image / label / regularization.
180			:param inputs_dict: inputs for loss function
181			"""
182
183			if name not in self.config["loss"]:
184			# loss config is not defined
185			logging.warning(
186			f"The configuration for loss {name} is not defined. "
187			f"Therefore it is not used."
188			)
189			return
190
191			loss_configs = self.config["loss"][name]
192			if not isinstance(loss_configs, list):
193			loss_configs = [loss_configs]
194
195			for loss_config in loss_configs:
196
197			if "weight" not in loss_config:
198			# default loss weight 1
199			logging.warning(
200			f"The weight for loss {name} is not defined."
201			f"Default weight = 1.0 is used."
202			)
203			loss_config["weight"] = 1.0
204
205			# build loss
206			weight = loss_config["weight"]
207
208			if weight == 0:
209			logging.warning(
210			f"The weight for loss {name} is zero." f"Loss is not used."
211			)
212			return
213
214			loss_layer: tf.keras.layers.Layer = REGISTRY.build_loss(
215			config=dict_without(d=loss_config, key="weight")
216			)
217			loss_value = loss_layer(**inputs_dict) / self.global_batch_size
218			weighted_loss = loss_value * weight
219
220			# add loss
221			self._model.add_loss(weighted_loss)
222
223			# add metric
224			self._model.add_metric(
225			loss_value, name=f"loss/{name}_{loss_layer.name}", aggregation="mean"
226			)
227			self._model.add_metric(
228			weighted_loss,
229			name=f"loss/{name}_{loss_layer.name}_weighted",
230			aggregation="mean",
231			)
232
233			@abstractmethod
234			def build_loss(self):
235			"""Build losses according to configs."""
236
237			def call(
238			self, inputs: Dict[str, tf.Tensor], training=None, mask=None
239			) -> Dict[str, tf.Tensor]:
240			"""
241			Call the self._model.
242
243			:param inputs: a dict of tensors.
244			:param training: training or not.
245			:param mask: maks for inputs.
246			:return:
247			"""
248			return self._model(inputs, training=training, mask=mask) # pragma: no cover
249
250			@abstractmethod
251			def postprocess(
252			self,
253			inputs: Dict[str, tf.Tensor],
254			outputs: Dict[str, tf.Tensor],
255			) -> Tuple[tf.Tensor, Dict]:
256			"""
257			Return a dict used for saving inputs and outputs.
258
259			:param inputs: dict of model inputs
260			:param outputs: dict of model outputs
261			:return: tuple, indices and a dict.
262			In the dict, each value is (tensor, normalize, on_label), where
263			- normalize = True if the tensor need to be normalized to [0, 1]
264			- on_label = True if the tensor depends on label
265			"""
266
267			def plot_model(self, output_dir: str):
268			"""
269			Save model structure in png.
270
271			:param output_dir: path to the output dir.
272			"""
273			logging.info(self._model.summary())
274			try:
275			tf.keras.utils.plot_model(
276			self._model,
277			to_file=os.path.join(output_dir, f"{self.name}.png"),
278			dpi=96,
279			show_shapes=True,
280			show_layer_names=True,
281			expand_nested=False,
282			)
283			except ImportError as err: # pragma: no cover
284			logging.error(
285			"Failed to plot model structure."
286			"Please check if graphviz is installed.\n"
287			"Error message is:"
288			f"{err}"
289			)
290
291			def log_tensor_stats(self, tensor: tf.Tensor, name: str):
292			"""
293			Log statistics of a given tensor.
294
295			:param tensor: tensor to monitor.
296			:param name: name of the tensor.
297			"""
298			flatten = tf.reshape(tensor, shape=(self.batch_size, -1))
299			self._model.add_metric(
300			tf.reduce_mean(flatten, axis=1),
301			name=f"metric/{name}_mean",
302			aggregation="mean",
303			)
304			self._model.add_metric(
305			tf.reduce_min(flatten, axis=1),
306			name=f"metric/{name}_min",
307			aggregation="min",
308			)
309			self._model.add_metric(
310			tf.reduce_max(flatten, axis=1),
311			name=f"metric/{name}_max",
312			aggregation="max",
313			)
314
315
316			@REGISTRY.register_model(name="ddf")
317			class DDFModel(RegistrationModel):
318			"""
319			A registration model predicts DDF.
320
321			When using global net as backbone,
322			the model predicts an affine transformation parameters,
323			and a DDF is calculated based on that.
324			"""
325
326			name = "DDFModel"
327
328			def _resize_interpolate(self, field, control_points):
329			resize = layer.ResizeCPTransform(control_points)
330			field = resize(field)
331
332			interpolate = layer.BSplines3DTransform(control_points, self.fixed_image_size)
333			field = interpolate(field)
334
335			return field
336
337			def build_model(self):
338			"""Build the model to be saved as self._model."""
339			# build inputs
340			self._inputs = self.build_inputs()
341			moving_image = self._inputs["moving_image"]
342			fixed_image = self._inputs["fixed_image"]
343
344			# build ddf
345			control_points = self.config["backbone"].pop("control_points", False)
346			backbone_inputs = self.concat_images(moving_image, fixed_image)
347			backbone = REGISTRY.build_backbone(
348			config=self.config["backbone"],
349			default_args=dict(
350			image_size=self.fixed_image_size,
351			out_channels=3,
352			out_kernel_initializer="zeros",
353			out_activation=None,
354			),
355			)
356
357			if isinstance(backbone, GlobalNet):
358			# (f_dim1, f_dim2, f_dim3, 3), (4, 3)
359			ddf, theta = backbone(inputs=backbone_inputs)
360			self._outputs = dict(ddf=ddf, theta=theta)
361			else:
362			# (f_dim1, f_dim2, f_dim3, 3)
363			ddf = backbone(inputs=backbone_inputs)
364			ddf = (
365			self._resize_interpolate(ddf, control_points) if control_points else ddf
366			)
367			self._outputs = dict(ddf=ddf)
368
369			# build outputs
370			self._warping = layer.Warping(fixed_image_size=self.fixed_image_size)
371			# (f_dim1, f_dim2, f_dim3, 3)
372			pred_fixed_image = self._warping(inputs=[ddf, moving_image])
373			self._outputs["pred_fixed_image"] = pred_fixed_image
374
375			if not self.labeled:
376			return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)
377
378			# (f_dim1, f_dim2, f_dim3, 3)
379			moving_label = self._inputs["moving_label"]
380			pred_fixed_label = self._warping(inputs=[ddf, moving_label])
381
382			self._outputs["pred_fixed_label"] = pred_fixed_label
383			return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)
384
385			def build_loss(self):
386			"""Build losses according to configs."""
387			fixed_image = self._inputs["fixed_image"]
388			moving_image = self._inputs["moving_image"]
389			ddf = self._outputs["ddf"]
390			pred_fixed_image = self._outputs["pred_fixed_image"]
391
392			# inputs
393			self.log_tensor_stats(tensor=moving_image, name="moving_image")
394			self.log_tensor_stats(tensor=fixed_image, name="fixed_image")
395
396			# ddf
397			self._build_loss(name="regularization", inputs_dict=dict(inputs=ddf))
398			self.log_tensor_stats(tensor=ddf, name="ddf")
399
400			# image
401			self._build_loss(
402			name="image", inputs_dict=dict(y_true=fixed_image, y_pred=pred_fixed_image)
403			)
404
405			# label
406			if self.labeled:
407			fixed_label = self._inputs["fixed_label"]
408			moving_label = self._inputs["moving_label"]
409			pred_fixed_label = self._outputs["pred_fixed_label"]
410			self._build_loss(
411			name="label",
412			inputs_dict=dict(y_true=fixed_label, y_pred=pred_fixed_label),
413			)
414			tre = compute_centroid_distance(
415			y_true=fixed_label, y_pred=pred_fixed_label, grid=self._warping.grid_ref
416			)
417			self._model.add_metric(tre, name="metric/TRE", aggregation="mean")
418			self.log_tensor_stats(tensor=moving_label, name="moving_label")
419			self.log_tensor_stats(tensor=fixed_label, name="fixed_label")
420
421			def postprocess(
422			self,
423			inputs: Dict[str, tf.Tensor],
424			outputs: Dict[str, tf.Tensor],
425			) -> Tuple[tf.Tensor, Dict]:
426			"""
427			Return a dict used for saving inputs and outputs.
428
429			:param inputs: dict of model inputs
430			:param outputs: dict of model outputs
431			:return: tuple, indices and a dict.
432			In the dict, each value is (tensor, normalize, on_label), where
433			- normalize = True if the tensor need to be normalized to [0, 1]
434			- on_label = True if the tensor depends on label
435			"""
436			indices = inputs["indices"]
437			processed = dict(
438			moving_image=(inputs["moving_image"], True, False),
439			fixed_image=(inputs["fixed_image"], True, False),
440			ddf=(outputs["ddf"], True, False),
441			pred_fixed_image=(outputs["pred_fixed_image"], True, False),
442			)
443
444			# save theta for affine model
445			if "theta" in outputs:
446			processed["theta"] = (outputs["theta"], None, None) # type: ignore
447
448			if not self.labeled:
449			return indices, processed
450
451			processed = {
452			**dict(
453			moving_label=(inputs["moving_label"], False, True),
454			fixed_label=(inputs["fixed_label"], False, True),
455			pred_fixed_label=(outputs["pred_fixed_label"], False, True),
456			),
457			**processed,
458			}
459
460			return indices, processed
461
462
463			@REGISTRY.register_model(name="dvf")
464			class DVFModel(DDFModel):
465			"""
466			A registration model predicts DVF.
467
468			DDF is calculated based on DVF.
469			"""
470
471			name = "DVFModel"
472
473			def build_model(self):
474			"""Build the model to be saved as self._model."""
475			# build inputs
476			self._inputs = self.build_inputs()
477			moving_image = self._inputs["moving_image"]
478			fixed_image = self._inputs["fixed_image"]
479			control_points = self.config["backbone"].pop("control_points", False)
480
481			# build ddf
482			backbone_inputs = self.concat_images(moving_image, fixed_image)
483			backbone = REGISTRY.build_backbone(
484			config=self.config["backbone"],
485			default_args=dict(
486			image_size=self.fixed_image_size,
487			out_channels=3,
488			out_kernel_initializer="zeros",
489			out_activation=None,
490			),
491			)
492			dvf = backbone(inputs=backbone_inputs)
493			dvf = self._resize_interpolate(dvf, control_points) if control_points else dvf
494			ddf = layer.IntDVF(fixed_image_size=self.fixed_image_size)(dvf)
495
496			# build outputs
497			self._warping = layer.Warping(fixed_image_size=self.fixed_image_size)
498			# (f_dim1, f_dim2, f_dim3, 3)
499			pred_fixed_image = self._warping(inputs=[ddf, moving_image])
500
501			self._outputs = dict(dvf=dvf, ddf=ddf, pred_fixed_image=pred_fixed_image)
502
503			if not self.labeled:
504			return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)
505
506			# (f_dim1, f_dim2, f_dim3, 3)
507			moving_label = self._inputs["moving_label"]
508			pred_fixed_label = self._warping(inputs=[ddf, moving_label])
509
510			self._outputs["pred_fixed_label"] = pred_fixed_label
511			return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)
512
513			def postprocess(
514			self,
515			inputs: Dict[str, tf.Tensor],
516			outputs: Dict[str, tf.Tensor],
517			) -> Tuple[tf.Tensor, Dict]:
518			"""
519			Return a dict used for saving inputs and outputs.
520
521			:param inputs: dict of model inputs
522			:param outputs: dict of model outputs
523			:return: tuple, indices and a dict.
524			In the dict, each value is (tensor, normalize, on_label), where
525			- normalize = True if the tensor need to be normalized to [0, 1]
526			- on_label = True if the tensor depends on label
527			"""
528			indices, processed = super().postprocess(inputs=inputs, outputs=outputs)
529			processed["dvf"] = (outputs["dvf"], True, False)
530			return indices, processed
531
532
533			@REGISTRY.register_model(name="conditional")
534			class ConditionalModel(RegistrationModel):
535			"""
536			A registration model predicts fixed image label without DDF or DVF.
537			"""
538
539			name = "ConditionalModel"
540
541			def build_model(self):
542			"""Build the model to be saved as self._model."""
543			assert self.labeled
544
545			# build inputs
546			self._inputs = self.build_inputs()
547			moving_image = self._inputs["moving_image"]
548			fixed_image = self._inputs["fixed_image"]
549			moving_label = self._inputs["moving_label"]
550
551			# build ddf
552			backbone_inputs = self.concat_images(moving_image, fixed_image, moving_label)
553			backbone = REGISTRY.build_backbone(
554			config=self.config["backbone"],
555			default_args=dict(
556			image_size=self.fixed_image_size,
557			out_channels=1,
558			out_kernel_initializer="glorot_uniform",
559			out_activation="sigmoid",
560			),
561			)
562			# (batch, f_dim1, f_dim2, f_dim3)
563			pred_fixed_label = backbone(inputs=backbone_inputs)
564			pred_fixed_label = tf.squeeze(pred_fixed_label, axis=4)
565
566			self._outputs = dict(pred_fixed_label=pred_fixed_label)
567			return tf.keras.Model(inputs=self._inputs, outputs=self._outputs)
568
569			def build_loss(self):
570			"""Build losses according to configs."""
571			fixed_label = self._inputs["fixed_label"]
572			pred_fixed_label = self._outputs["pred_fixed_label"]
573
574			self._build_loss(
575			name="label",
576			inputs_dict=dict(y_true=fixed_label, y_pred=pred_fixed_label),
577			)
578
579			def postprocess(
580			self,
581			inputs: Dict[str, tf.Tensor],
582			outputs: Dict[str, tf.Tensor],
583			) -> Tuple[tf.Tensor, Dict]:
584			"""
585			Return a dict used for saving inputs and outputs.
586
587			:param inputs: dict of model inputs
588			:param outputs: dict of model outputs
589			:return: tuple, indices and a dict.
590			In the dict, each value is (tensor, normalize, on_label), where
591			- normalize = True if the tensor need to be normalized to [0, 1]
592			- on_label = True if the tensor depends on label
593			"""
594			indices = inputs["indices"]
595			processed = dict(
596			moving_image=(inputs["moving_image"], True, False),
597			fixed_image=(inputs["fixed_image"], True, False),
598			pred_fixed_label=(outputs["pred_fixed_label"], True, True),
599			moving_label=(inputs["moving_label"], False, True),
600			fixed_label=(inputs["fixed_label"], False, True),
601			)
602
603			return indices, processed
604

DeepRegNet / DeepReg

Pull Request — main (#713)

RegistrationModel.log_tensor_stats() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like