#include <ObjectiveFunction.h>

Inheritance diagram for mirtk::ObjectiveFunction:

Collaboration diagram for mirtk::ObjectiveFunction:

Public Member Functions
virtual void	AddConstraintGradient (double dx, double step=.0, bool sgn_chg=nullptr)

virtual void	DataFidelityGradient (double dx, double step=.0, bool sgn_chg=nullptr)

virtual double	Evaluate (double dx=NULL, double step=.0, bool sgn_chg=NULL)

virtual double	Get (int i) const =0

virtual void	Get (double *x) const =0

virtual void	Gradient (double dx, double step=.0, bool sgn_chg=nullptr)=0

virtual double	GradientNorm (const double *dx) const =0

virtual void	GradientStep (const double *dx, double &min, double &max) const

virtual int	NumberOfDOFs () const =0

virtual void	Put (const double *x)=0

virtual double	Step (double *dx)=0

virtual void	Update (bool gradient=true)

virtual bool	Upgrade ()

virtual double	Value ()=0
	Evaluate objective function value.

virtual	~ObjectiveFunction ()=0
	Destructor.

Public Member Functions inherited from mirtk::Observable
void	AddObserver (Observer &)
	Add observer.

void	Broadcast (Event, const void *=NULL)
	Broadcast event to observers.

void	ClearObservers ()
	Delete all observers.

void	DeleteObserver (Observer &)
	Delete observer.

void	NotifyObservers (Event, const void *=NULL)
	Notify all observers about given event if this object has changed.

int	NumberOfObservers () const
	Number of current observers.

virtual	~Observable ()
	Destructor.

Public Member Functions inherited from mirtk::Object
virtual const char *	NameOfClass () const =0
	Get name of class, which this object is an instance of.

virtual ParameterList	Parameter () const
	Get parameter name/value pairs.

bool	Parameter (const ParameterList &)
	Set parameters from name/value pairs.

virtual bool	Set (const char name, const char value)

virtual	~Object ()
	Destructor.

Protected Member Functions
	mirtkAttributeMacro (double, StepLength)
	Default step length used for gradient approximation.

Protected Member Functions inherited from mirtk::Observable
	Observable ()
	Default constructor.

	Observable (const Observable &)
	Copy constructor.

Observable &	operator= (const Observable &)
	Assignment operator.

Protected Member Functions inherited from mirtk::Object
template<typename... Args>
void	Throw (ErrorType err, const char *func, Args... args) const

Additional Inherited Members
Static Public Member Functions inherited from mirtk::Object
static const char *	NameOfType ()
	Get name of this class type.

Static Protected Member Functions inherited from mirtk::Object
template<typename... Args>
static void	ThrowStatic (ErrorType err, const char cls, const char func, Args... args)

Detailed Description

Interface for objective function of an optimization problem.

An objective function can be minimized/maximized using a specialization of the irtkLocalOptimizer base class.

Definition at line 35 of file ObjectiveFunction.h.

Member Function Documentation

§ AddConstraintGradient()

void mirtk::ObjectiveFunction::AddConstraintGradient	(	double *	dx,
		double	step = `.0`,
		bool *	sgn_chg = `nullptr`
	)

inlinevirtual

Add model constraint gradient of objective function w.r.t its DoFs

Parameters

[in]	step	Step length for finite differences.
[out]	dx	Gradient of objective function, e.g., either initialised to zero or the data fidelity gradient computed beforehand using DataFidelityGradient.
[out]	sgn_chg	Whether function parameter value is allowed to change sign when stepping along the computed gradient.

Reimplemented in mirtk::RegistrationEnergy.

Definition at line 226 of file ObjectiveFunction.h.

§ DataFidelityGradient()

void mirtk::ObjectiveFunction::DataFidelityGradient	(	double *	dx,
		double	step = `.0`,
		bool *	sgn_chg = `nullptr`
	)

inlinevirtual

Evaluate data fidelity gradient of objective function w.r.t its DoFs

Parameters

[in]	step	Step length for finite differences.
[out]	dx	Data fidelity gradient of objective function.
[out]	sgn_chg	Whether function parameter value is allowed to change sign when stepping along the computed gradient.

Reimplemented in mirtk::RegistrationEnergy.

Definition at line 219 of file ObjectiveFunction.h.

§ Evaluate()

double mirtk::ObjectiveFunction::Evaluate	(	double *	dx = `NULL`,
		double	step = `.0`,
		bool *	sgn_chg = `NULL`
	)

inlinevirtual

Evaluate objective function

This function first updates the internal state of the function object if required due to a previous change of the function parameters and then evaluates the current function value. If dx is not NULL, the gradient of the objective function is computed as well.

Parameters

[in]	step	Step length for finite differences.
[out]	dx	Gradient of objective function. If `NULL`, only the function value is computed.
[out]	sgn_chg	Whether function parameter value is allowed to change sign when stepping along the computed gradient. Ignord if `dx` is `NULL`.

Returns: Value of the objective function.

Reimplemented in mirtk::DeformableSurfaceModel, and mirtk::RegistrationEnergy.

Definition at line 232 of file ObjectiveFunction.h.

§ Get() [1/2]

virtual double mirtk::ObjectiveFunction::Get ( int i ) const

pure virtual

Get function parameter value

Parameters

[in] i Function parameter (DoF) index.

Returns: Value of specified function parameter (DoF).

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ Get() [2/2]

virtual void mirtk::ObjectiveFunction::Get ( double * x ) const

pure virtual

Get function parameter values

This function can be used to store a backup of the current funtion parameter values before an update such that these can be restored using the Put member function if the update did not result in the desired change of the overall objective function value.

Parameters

[in] x Function parameter (DoF) values.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ Gradient()

virtual void mirtk::ObjectiveFunction::Gradient	(	double *	dx,
		double	step = `.0`,
		bool *	sgn_chg = `nullptr`
	)

pure virtual

Evaluate gradient of objective function w.r.t its DoFs

When the data fidelity and/or gradient of model constraints need to be evaluated separately, use the following code instead:

double *gradient;        // allocate with number of DoFs
ObjectiveFunction *func; // pointer to objective function instance
func->DataFideltiyGradient(gradient);
func->AddConstraintGradient(gradient);

Parameters

[in]	step	Step length for finite differences.
[out]	dx	Gradient of objective function.
[out]	sgn_chg	Whether function parameter value is allowed to change sign when stepping along the computed gradient.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ GradientNorm()

virtual double mirtk::ObjectiveFunction::GradientNorm ( const double * dx ) const

pure virtual

Compute norm of gradient of objective function

This norm is used to define a unit for the step length used by gradient descent methods. It is, for example, the maximum absolute value norm for linear transformations and the maximum control point displacement for FFDs. The computation of the norm may be done after conjugating the gradient vector obtained using the Gradient member function.

Parameters

[in] dx Gradient of objective function.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ GradientStep()

void mirtk::ObjectiveFunction::GradientStep	(	const double *	dx,
		double &	min,
		double &	max
	)		const

inlinevirtual

Adjust step length range

Parameters

[in]	dx	Gradient of objective function.
[in,out]	min	Minimum step length.
[in,out]	max	Maximum step length.

Reimplemented in mirtk::DeformableSurfaceModel, and mirtk::RegistrationEnergy.

Definition at line 213 of file ObjectiveFunction.h.

§ NumberOfDOFs()

virtual int mirtk::ObjectiveFunction::NumberOfDOFs ( ) const

pure virtual

Get number of DoFs

Returns: Number of free function parameters.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ Put()

virtual void mirtk::ObjectiveFunction::Put ( const double * x )

pure virtual

Set function parameter values

This is function can be used to set the parameters of the objective function to particular values. In particular, it can be used to restore the function parameters after a failed incremental update which did not result in the desired improvement.

Parameters

[in] x Function parameter (DoF) values.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ Step()

virtual double mirtk::ObjectiveFunction::Step ( double * dx )

pure virtual

Add change (i.e., scaled gradient) to each parameter value

This function updates each DoF of the objective function given a vector of corresponding changes, i.e., the computed gradient of the objective function w.r.t. these parameters or a desired change computed otherwise.

Parameters

[in,out] dx Change of each function parameter (DoF) as computed by the Gradient member function and scaled by a chosen step length. The change of a parameter may be modified by this function in order to satisfy the hard constraints (if any).

Returns: Maximum change of function parameter.

Implemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

§ Update()

void mirtk::ObjectiveFunction::Update ( bool gradient = true )

inlinevirtual

Update internal state after change of parameters

Parameters

[in] gradient Update also internal state required for evaluation of gradient of objective function.

Reimplemented in mirtk::DeformableSurfaceModel, mirtk::RegistrationEnergy, and mirtk::TransformationApproximationError.

Definition at line 202 of file ObjectiveFunction.h.

§ Upgrade()

bool mirtk::ObjectiveFunction::Upgrade ( )

inlinevirtual

Update objective function after convergence

This function may be called by the optimizer after the optimization of the current objective function has converged or the improvement is insignificant. It allows the objective function to modify itself considering the current estimate of the parameters (DoFs). For example, the fiducial registration error term of a registration energy function could update the point correspondences.

Returns: Whether the objective function has changed.

Reimplemented in mirtk::DeformableSurfaceModel, and mirtk::RegistrationEnergy.

Definition at line 207 of file ObjectiveFunction.h.

The documentation for this class was generated from the following file:

ObjectiveFunction.h

Public Member Functions

Protected Member Functions

Additional Inherited Members

Detailed Description

Member Function Documentation

§ AddConstraintGradient()

§ DataFidelityGradient()

§ Evaluate()

§ Get() [1/2]

§ Get() [2/2]

§ Gradient()

§ GradientNorm()

§ GradientStep()

§ NumberOfDOFs()

§ Put()

§ Step()

§ Update()

§ Upgrade()