 |
ImFusion SDK 4.3
|
|
ImFusion SDK 4.3
|
#include <ImFusion/CT/Solver/Functional.h>
Abstract functional interface for optimization problems. More...
Inheritance diagram for Functional:Abstract functional interface for optimization problems.
Represents a (sufficiently smooth) functional on images, i.e. a (sufficiently smooth) function which takes as input images/volumes and returns a real value as output. The prototypical example of such a functional is the sum of squares of intensities in the image/volume. Functionals in CT typically arise in connection to linear residuals, i.e. if \( f\) is a functional, the derived quantity \( f( Ax - b) \) measures something like the residual of a CT reconstruction \( x \) given projections \( b \). If a linear residual is provided to this class, we pull-back values, gradients and the hessians by this residual, i.e. we provide the result for the functional \( x \mapsto f(Ax - b)\).
Public Types | |
| using | ExprBase = ImageMath::ExprBase |
Public Member Functions | |
| Functional (std::unique_ptr< LinearResidual > residual, bool hessianRequiresResidualEval=true) | |
| Constructor with residual. This residual can be nullptr, which is taken as shorthand for the identity. | |
| Functional (std::unique_ptr< LinearOperator > A, SharedImageSet *b, bool hessianRequiresResidualEval=true) | |
| Constructor which implicitly constructs a residual that owns the LinearOperator. | |
| virtual | ~Functional ()=0 |
| Destructor. | |
| float | eval (const SharedImageSet &in) const |
| Evaluates the functional \(f(in)\). | |
| Status | gradient (const SharedImageSet &in, SharedImageSet &out, const Selection *sel=nullptr, const ExprBase *exprOut=nullptr) const |
| Evaluates the gradient \(\nabla f(in)\). | |
| template<typename ImplName> | |
| Status | gradient (const SharedImageSet &in, SharedImageSet &out, const Selection *sel, const ImageMath::ExprBaseTpl< ImplName > *exprOut) |
| Convenience function which wraps exprOut in sharedPolyWrapper. | |
| std::unique_ptr< LinearOperator > | hessian (const SharedImageSet &in) const |
| Evaluates the hessian \(H_f(in)\). | |
| const LinearResidual * | residual () const |
| Const getter for m_residual. | |
| void | configure (const Properties *p) override |
| Configure this object instance by de-serializing the given Properties. | |
| void | configuration (Properties *p) const override |
| Serialize the current object configuration into the given Properties object. | |
| Public Member Functions inherited from Configurable | |
| virtual void | configureDefaults () |
| Retrieve the properties of this object, replaces values with their defaults and sets it again. | |
| void | registerParameter (ParameterBase *param) |
| Register the given Parameter or SubProperty, so that it will be configured during configure()/configuration(). | |
| void | unregisterParameter (const ParameterBase *param) |
| Remove the given Parameter or SubProperty from the list of registered parameters. | |
| Configurable (const Configurable &rhs) | |
| Configurable (Configurable &&rhs) noexcept | |
| Configurable & | operator= (const Configurable &) |
| Configurable & | operator= (Configurable &&) noexcept |
Protected Member Functions | |
| virtual float | evalImpl (const SharedImageSet &in) const =0 |
| Evaluation of this functional (without residual) | |
| virtual Status | gradientImpl (const SharedImageSet &in, SharedImageSet &out) const =0 |
| Gradient of this functional (without residual). Should be selection aware. | |
| virtual std::unique_ptr< LinearOperator > | hessianImpl (const SharedImageSet &in) const =0 |
| Hessian of this functional (without residual) | |
Protected Attributes | |
| std::unique_ptr< LinearResidual > | m_residual |
| optional residual. | |
| const bool | m_hessianRequiresResidualEval = true |
| if set to false the result of hessianImpl is expected not to depend on the input | |
| detail::SparseSharedImageSet | m_residualTempSparseSharedImageSet |
| Sparse tmp image. | |
| Protected Attributes inherited from Configurable | |
| std::vector< Param > | m_params |
| List of all registered Parameter and SubProperty instances. | |
Additional Inherited Members | |
| Public Attributes inherited from Configurable | |
| Signal | signalParametersChanged |
Emitted whenever one of the registered Parameters' or SubPropertys' signalValueChanged signal was emitted. | |
| float eval | ( | const SharedImageSet & | in | ) | const |
Evaluates the functional \(f(in)\).
| [in] | in | SharedImageSet at which the functional is evaluated. |
| Status gradient | ( | const SharedImageSet & | in, |
| SharedImageSet & | out, | ||
| const Selection * | sel = nullptr, | ||
| const ExprBase * | exprOut = nullptr ) const |
Evaluates the gradient \(\nabla f(in)\).
| [in] | in | SharedImageSet at which the gradient is evaluated. |
| [out] | out | gradient at the given value. |
| [in] | sel | Selection on the range of the residual. Can be used for evaluating subsets in case a residual with an operator is set. Otherwise sel MUST be nullptr. |
| [in] | exprOut | If set, this will be applied to the result of the gradient. This can be used to implicitly perform a gradient step for instance. |
| std::unique_ptr< LinearOperator > hessian | ( | const SharedImageSet & | in | ) | const |
Evaluates the hessian \(H_f(in)\).
| [in] | in | SharedImageSet at which the hessian is evaluated. |
|
overridevirtual |
Configure this object instance by de-serializing the given Properties.
The default implementation will do so automatically for all registered Parameter and SubProperty instances.
Reimplemented from Configurable.
|
overridevirtual |
Serialize the current object configuration into the given Properties object.
The default implementation will do so automatically for all registered Parameter and SubProperty instances.
Reimplemented from Configurable.
|
protectedpure virtual |
Evaluation of this functional (without residual)
Implemented in AnisotropicTV, IsotropicTV, L2Pow2, and WeightedL2Pow2< OperandT >.
|
protectedpure virtual |
Gradient of this functional (without residual). Should be selection aware.
Implemented in AnisotropicTV, IsotropicTV, L2Pow2, and WeightedL2Pow2< OperandT >.
|
protectedpure virtual |
Hessian of this functional (without residual)
Implemented in AnisotropicTV, IsotropicTV, L2Pow2, and WeightedL2Pow2< OperandT >.
