Module deformation_inversion_layer.interface

Interface definitions

"""Interface definitions"""

from abc import abstractmethod
from typing import Protocol, Sequence

from torch import Tensor


class Interpolator(Protocol):
    """Interpolates values on regular grid in voxel coordinates"""

    @abstractmethod
    def __call__(self, volume: Tensor, coordinates: Tensor) -> Tensor:
        """Interpolate

        Args:
            volume: Volume to be interpolated with shape
                (batch_size, *channel_dims, dim_1, ..., dim_{n_dims}). Dimension
                order is the same as the coordinate order of the coordinates
            coordinates: Interpolation coordinates with shape
                (batch_size, n_dims, *target_shape)

        Returns:
            Interpolated volume with shape (batch_size, *channel_dims, *target_shape)
        """


class FixedPointFunction(Protocol):
    """Protocol for fixed point functions"""

    def __call__(
        self,
        iteration_input: Tensor,
        output_buffer: Tensor,
    ) -> None:
        """Call a fixed point function

        Args:
            iteration_input: Input to the fixed point function
            output_buffer: Output should be stored in-place to this Tensor (same
                shape as the input)
        """


class FixedPointSolver(Protocol):
    """Protocol for fixed point solvers"""

    @abstractmethod
    def solve(
        self,
        fixed_point_function: FixedPointFunction,
        initial_value: Tensor,
    ) -> Tensor:
        """Solve a fixed point problem

        Args:
            fixed_point_function: Function to be iterated until convergence
            initial_value: Initial iteration value

        Returns:
            Solution of the fixed point iteration
        """


class FixedPointStopCriterion(Protocol):
    """Protocol for fixed point iteration stopping criterions"""

    @abstractmethod
    def should_stop(
        self,
        current_iteration: Tensor,
        previous_iterations: Sequence[Tensor],
        n_earlier_iterations: int,
    ) -> bool:
        """Return whether iterating should be continued at beginning of an iteration

        Args:
            current_iteration: Current output of the fixed point iteration. For
                n_earlier_iterations == 0 this equals the initial guess.
            previous_iterations: Previous outputs of the fixed point iteration,
                starting from the most recent one. length of this list may
                depend on the fixed point iteration solver and is always 0 for
                the n_earlier_iterations == 0.
            n_earlier_iterations: Number of calls made to the fixed point function
                before the next iteration.

        Returns:
            Whether the iteration should be stopped
        """

Classes

class FixedPointFunction (*args, **kwargs)

Protocol for fixed point functions

Expand source code

class FixedPointFunction(Protocol):
    """Protocol for fixed point functions"""

    def __call__(
        self,
        iteration_input: Tensor,
        output_buffer: Tensor,
    ) -> None:
        """Call a fixed point function

        Args:
            iteration_input: Input to the fixed point function
            output_buffer: Output should be stored in-place to this Tensor (same
                shape as the input)
        """

Ancestors

• typing.Protocol
• typing.Generic

class FixedPointSolver (*args, **kwargs)

Protocol for fixed point solvers

Expand source code

class FixedPointSolver(Protocol):
    """Protocol for fixed point solvers"""

    @abstractmethod
    def solve(
        self,
        fixed_point_function: FixedPointFunction,
        initial_value: Tensor,
    ) -> Tensor:
        """Solve a fixed point problem

        Args:
            fixed_point_function: Function to be iterated until convergence
            initial_value: Initial iteration value

        Returns:
            Solution of the fixed point iteration
        """

Ancestors

• typing.Protocol
• typing.Generic

Subclasses

• AndersonSolver
• EmptySolver
• NaiveSolver

Methods

def solve(self, fixed_point_function: FixedPointFunction, initial_value: torch.Tensor) ‑> torch.Tensor

Solve a fixed point problem

Args

fixed_point_function

Function to be iterated until convergence

initial_value

Initial iteration value

Returns

Solution of the fixed point iteration

Expand source code

@abstractmethod
def solve(
    self,
    fixed_point_function: FixedPointFunction,
    initial_value: Tensor,
) -> Tensor:
    """Solve a fixed point problem

    Args:
        fixed_point_function: Function to be iterated until convergence
        initial_value: Initial iteration value

    Returns:
        Solution of the fixed point iteration
    """

class FixedPointStopCriterion (*args, **kwargs)

Protocol for fixed point iteration stopping criterions

Expand source code

class FixedPointStopCriterion(Protocol):
    """Protocol for fixed point iteration stopping criterions"""

    @abstractmethod
    def should_stop(
        self,
        current_iteration: Tensor,
        previous_iterations: Sequence[Tensor],
        n_earlier_iterations: int,
    ) -> bool:
        """Return whether iterating should be continued at beginning of an iteration

        Args:
            current_iteration: Current output of the fixed point iteration. For
                n_earlier_iterations == 0 this equals the initial guess.
            previous_iterations: Previous outputs of the fixed point iteration,
                starting from the most recent one. length of this list may
                depend on the fixed point iteration solver and is always 0 for
                the n_earlier_iterations == 0.
            n_earlier_iterations: Number of calls made to the fixed point function
                before the next iteration.

        Returns:
            Whether the iteration should be stopped
        """

Ancestors

• typing.Protocol
• typing.Generic

Subclasses

• BaseCountingStopCriterion

Methods

def should_stop(self, current_iteration: torch.Tensor, previous_iterations: Sequence[torch.Tensor], n_earlier_iterations: int) ‑> bool

Return whether iterating should be continued at beginning of an iteration

Args

current_iteration

Current output of the fixed point iteration. For n_earlier_iterations == 0 this equals the initial guess.

previous_iterations

Previous outputs of the fixed point iteration, starting from the most recent one. length of this list may depend on the fixed point iteration solver and is always 0 for the n_earlier_iterations == 0.

n_earlier_iterations

Number of calls made to the fixed point function before the next iteration.

Returns

Whether the iteration should be stopped

Expand source code

@abstractmethod
def should_stop(
    self,
    current_iteration: Tensor,
    previous_iterations: Sequence[Tensor],
    n_earlier_iterations: int,
) -> bool:
    """Return whether iterating should be continued at beginning of an iteration

    Args:
        current_iteration: Current output of the fixed point iteration. For
            n_earlier_iterations == 0 this equals the initial guess.
        previous_iterations: Previous outputs of the fixed point iteration,
            starting from the most recent one. length of this list may
            depend on the fixed point iteration solver and is always 0 for
            the n_earlier_iterations == 0.
        n_earlier_iterations: Number of calls made to the fixed point function
            before the next iteration.

    Returns:
        Whether the iteration should be stopped
    """

class Interpolator (*args, **kwargs)

Interpolates values on regular grid in voxel coordinates

Expand source code

class Interpolator(Protocol):
    """Interpolates values on regular grid in voxel coordinates"""

    @abstractmethod
    def __call__(self, volume: Tensor, coordinates: Tensor) -> Tensor:
        """Interpolate

        Args:
            volume: Volume to be interpolated with shape
                (batch_size, *channel_dims, dim_1, ..., dim_{n_dims}). Dimension
                order is the same as the coordinate order of the coordinates
            coordinates: Interpolation coordinates with shape
                (batch_size, n_dims, *target_shape)

        Returns:
            Interpolated volume with shape (batch_size, *channel_dims, *target_shape)
        """

Ancestors

• typing.Protocol
• typing.Generic

Subclasses

• BicubicInterpolator
• LinearInterpolator
• NearestInterpolator