DirectGraphicalModels::CEdgeModelPotts Class Reference

Potts Edge Model for dense graphical models. More...

#include <EdgeModelPotts.h>

Inheritance diagram for DirectGraphicalModels::CEdgeModelPotts:

Collaboration diagram for DirectGraphicalModels::CEdgeModelPotts:

Public Member Functions
	CEdgeModelPotts (const Mat &features, float weight=1.0f, const std::function< void(const Mat &src, Mat &dst)> &semiMetricFunction={}, bool perPixelNormalization=true)
	Constructor. More...
virtual	~CEdgeModelPotts (void)=default
void	apply (const Mat &src, Mat &dst) const override
	Applies an edge model to the node potentials of a dense graph. More...
Public Member Functions inherited from DirectGraphicalModels::IEdgeModel
	IEdgeModel (void)=default
	IEdgeModel (const IEdgeModel &)=delete
virtual	~IEdgeModel (void)=default
const IEdgeModel &	operator= (const IEdgeModel &)=delete

Private Attributes
std::unique_ptr< CPermutohedral >	m_pLattice
	Pointer to the permutohedral lattice. More...
float	m_weight
	The weighting parameter. More...
Mat	m_norm
	Array with normalization factors. More...
std::function< void(const Mat &src, Mat &dst)>	m_function
	The semi-metric function. More...

Detailed Description

Potts Edge Model for dense graphical models.

This class implements Potts edge potential model in the fully connected CRF. The implementation is based on Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials paper.

Author

Sergey G. Kosov, serge.nosp@m.y.ko.nosp@m.sov@p.nosp@m.roje.nosp@m.ct-10.nosp@m..de

Definition at line 17 of file EdgeModelPotts.h.

Constructor & Destructor Documentation

CEdgeModelPotts()

DirectGraphicalModels::CEdgeModelPotts::CEdgeModelPotts	(	const Mat &	features,
		float	weight = 1.0f,
		const std::function< void(const Mat &src, Mat &dst)> &	semiMetricFunction = {},
		bool	perPixelNormalization = true
	)

Constructor.

This constucts a new edge potentials model, which is in general training-data-independent, but still contrast-sensitive model (see PhD Thesis Multi-Layer Conditional Random Fields for Revealing Unobserved Entities p. 60 for more details). Test-data-dependency is provided via features argument. However this model might be ''trained'' by using a semimetric function provided via semiMetricFunction argument.

Parameters

features	The set of features which correspond to the nodes of the dense graphical model: Mat(size: nNodes x nFeatures; type: CV_32FC1)
weight	The weighting parameter (default value is 1)
semiMetricFunction	Reference to a semi-metric function, which arguments src and dst are: Mat(size: 1 x nFeatures; type: CV_32FC1). This function when provided will be called for every node potential in the apply() method.
perPixelNormalization	Flag indicating whether er-pixel normalization should be used during applying the edge model.

Definition at line 5 of file EdgeModelPotts.cpp.

~CEdgeModelPotts()

virtual DirectGraphicalModels::CEdgeModelPotts::~CEdgeModelPotts ( void  )

virtualdefault

Member Function Documentation

apply()

void DirectGraphicalModels::CEdgeModelPotts::apply ( const Mat &  src, Mat &  dst  ) const

overridevirtual

Applies an edge model to the node potentials of a dense graph.

This function subsequently (in terms of multiple iterations and multiple models) applies an edge model derived from this class to the node potentials provided via src argument and stores the result into the dst.

Parameters

[in]	src	The dense graph node potentials in form Mat(size: nNodes x nStates; type: CV_32FC1)
[out]	dst	The reference to the container for resulting node potentials. Resulting matrix will be the same size and type as the input one: Mat(size: nNodes x nStates; type: CV_32FC1)

Implements DirectGraphicalModels::IEdgeModel.

Definition at line 28 of file EdgeModelPotts.cpp.

Member Data Documentation

m_function

std::function<void(const Mat &src, Mat &dst)> DirectGraphicalModels::CEdgeModelPotts::m_function

private

The semi-metric function.

Definition at line 41 of file EdgeModelPotts.h.

m_norm

Mat DirectGraphicalModels::CEdgeModelPotts::m_norm

private

Array with normalization factors.

Definition at line 40 of file EdgeModelPotts.h.

m_pLattice

std::unique_ptr<CPermutohedral> DirectGraphicalModels::CEdgeModelPotts::m_pLattice

private

Pointer to the permutohedral lattice.

Definition at line 38 of file EdgeModelPotts.h.

m_weight

float DirectGraphicalModels::CEdgeModelPotts::m_weight

private

The weighting parameter.

Definition at line 39 of file EdgeModelPotts.h.

---

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

• D:/Projects/DGM/modules/DGM/EdgeModelPotts.h
• D:/Projects/DGM/modules/DGM/EdgeModelPotts.cpp