Base abstract class for link (inter-layer edge) potentials training. More...

#include <TrainLink.h>

Inheritance diagram for DirectGraphicalModels::CTrainLink:

Collaboration diagram for DirectGraphicalModels::CTrainLink:

Public Member Functions
	CTrainLink (byte nStatesBase, byte nStatesOccl, word nFeatures)
	Constructor. More...

virtual	~CTrainLink (void)=default

void	addFeatureVec (const Mat &featureVectors, const Mat &gtb, const Mat &gto)
	Adds a block of new feature vectors. More...

void	addFeatureVec (const vec_mat_t &featureVectors, const Mat &gtb, const Mat &gto)
	Adds a block of new feature vectors. More...

virtual void	addFeatureVec (const Mat &featureVector, byte gtb, byte gto)=0
	Adds a feature vector. More...

virtual void	train (bool doClean=false)
	Random model training. More...

Mat	getLinkPotentials (const Mat &featureVector, float weight=1.0f) const
	Returns the link potential, based on the feature vector. More...

Public Member Functions inherited from DirectGraphicalModels::ITrain
	ITrain (byte nStates, word nFeatures)
	Constructor. More...

virtual	~ITrain (void)=default

word	getNumFeatures (void) const
	Returns number of features. More...

Public Member Functions inherited from DirectGraphicalModels::CBaseRandomModel
	CBaseRandomModel (byte nStates)
	Constructor. More...

virtual	~CBaseRandomModel (void)

virtual void	reset (void)=0
	Resets class variables. More...

virtual void	save (const std::string &path, const std::string &name=std::string(), short idx=-1) const
	Saves the training data. More...

virtual void	load (const std::string &path, const std::string &name=std::string(), short idx=-1)
	Loads the training data. More...

byte	getNumStates (void) const
	Returns number of states (classes) More...

Protected Member Functions
virtual Mat	calculateLinkPotentials (const Mat &featureVector) const =0
	Calculates the link potential, based on the feature vector. More...

Protected Member Functions inherited from DirectGraphicalModels::CBaseRandomModel
virtual void	saveFile (FILE *pFile) const =0
	Saves the random model into the file. More...

virtual void	loadFile (FILE *pFile)=0
	Loads the random model from the file. More...

std::string	generateFileName (const std::string &path, const std::string &name, short idx) const
	Generates name of the data file for storing random model parameters. More...

Protected Attributes
byte	m_nStatesBase
	Number of states (classes) at the base layer of ML-CRF. More...

byte	m_nStatesOccl
	Number of states (classes) at the occlusion layerts of ML-CRF. More...

Protected Attributes inherited from DirectGraphicalModels::CBaseRandomModel
byte	m_nStates
	The number of states (classes) More...

Detailed Description

Base abstract class for link (inter-layer edge) potentials training.

The link potentials (or occlusion potentials) are used in the multi-layer CRF model. Refer to
Labeling of Partially Occluded Regions via the Multi-Layer CRF paper for more details.

Author: Dr. Sergey 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 TrainLink.h.

Constructor & Destructor Documentation

◆ CTrainLink()

DirectGraphicalModels::CTrainLink::CTrainLink	(	byte	nStatesBase,
		byte	nStatesOccl,
		word	nFeatures
	)

inline

Constructor.

Parameters

nStatesBase	Number of states (classes) for the base layer of the graphical model
nStatesOccl	Number of states (classes) for the occlusion layer of the graphical model
nFeatures	Number of features

Definition at line 26 of file TrainLink.h.

◆ ~CTrainLink()

virtual DirectGraphicalModels::CTrainLink::~CTrainLink ( void )

virtualdefault

Member Function Documentation

◆ addFeatureVec() [1/3]

void DirectGraphicalModels::CTrainLink::addFeatureVec	(	const Mat &	featureVectors,
		const Mat &	gtb,
		const Mat &	gto
	)

Adds a block of new feature vectors.

Used to add multiple featureVectors, corresponding to the ground-truth states (classes) gtb and gto for training

Parameters

featureVectors	Multi-channel matrix, each element of which is a multi-dimensinal point: Mat(type: CV_8UC<nFeatures>)
gtb	Matrix, each element of which is a ground-truth state (class), corresponding to the base layer
gto	Matrix, each element of which is a ground-truth state (class), corresponding to the occlusion layer

Definition at line 6 of file TrainLink.cpp.

Here is the call graph for this function:

◆ addFeatureVec() [2/3]

void DirectGraphicalModels::CTrainLink::addFeatureVec	(	const vec_mat_t &	featureVectors,
		const Mat &	gtb,
		const Mat &	gto
	)

Adds a block of new feature vectors.

Used to add multiple featureVectors, corresponding to the ground-truth states (classes) gtb and gto for training

Parameters

featureVectors	Vector of size nFeatures, each element of which is a single feature - image: Mat(type: CV_8UC1)
gtb	Matrix, each element of which is a ground-truth state (class), corresponding to the base layer
gto	Matrix, each element of which is a ground-truth state (class), corresponding to the occlusion layer

Definition at line 12 of file TrainLink.cpp.

Here is the call graph for this function:

◆ addFeatureVec() [3/3]

virtual void DirectGraphicalModels::CTrainLink::addFeatureVec	(	const Mat &	featureVector,
		byte	gtb,
		byte	gto
	)

pure virtual

Adds a feature vector.

Used to add featureVector, corresponding to the ground-truth states (classes) gtb and gto for training. Here the couple {gtb, gto} corresponds to the nodes from base aod occlusion layers.

Parameters

featureVector	Multi-dimensinal point: Mat(size: nFeatures x 1; type: CV_8UC1), corresponding to both nodes of the link.
gtb	The ground-truth state (class) of the first node of the edge, corresponding to the base layer
gto	The ground-truth state (class) of the second node of the edge, corresponding to the occlusion layer

Implemented in DirectGraphicalModels::CTrainLinkNested< Trainer >.

◆ calculateLinkPotentials()

virtual Mat DirectGraphicalModels::CTrainLink::calculateLinkPotentials ( const Mat & featureVector ) const

protectedpure virtual

Calculates the link potential, based on the feature vector.

This function calculates the potentials of an edge, described with the sample featureVector, correspondig to the both nodes defining that edge. The resulting potentials of the two nodes being in each possible state (one class from occlusion layer occlusion another class from the base layer), are united in the edge potential matrix:

\[edgePot[nStates][nStates] = f(\textbf{f}[nFeatures])\]

Functions \( f \) must be implemented in derived classes.

Parameters

featureVector Multi-dimensinal point \(\textbf{f}\): Mat(size: nFeatures x 1; type: CV_{XX}C1), corresponding to both nodes of the link

Returns: The edge potential matrix: Mat(size: nStates x nStates; type: CV_32FC1)

Implemented in DirectGraphicalModels::CTrainLinkNested< Trainer >.

Here is the caller graph for this function:

◆ getLinkPotentials()

Mat DirectGraphicalModels::CTrainLink::getLinkPotentials	(	const Mat &	featureVector,
		float	weight = `1.0f`
	)		const

Returns the link potential, based on the feature vector.

This function calls calculateLinkPotentials() function, which should be implemented in derived classes. After that, the resulting edge potential is powered by parameter weight.

Parameters

featureVector	Multi-dimensinal point \(\textbf{f}\): Mat(size: nFeatures x 1; type: CV_{XX}C1), corresponding to both nodes of the link
weight	The weighting parameter

Returns: Edge potentials on success: Mat(size: nStates x nStates; type: CV_32FC1)

Definition at line 18 of file TrainLink.cpp.

Here is the call graph for this function:

◆ train()

virtual void DirectGraphicalModels::CTrainLink::train ( bool doClean = false )

inlinevirtual

Random model training.

Auxilary function for training - some derived classes may use this function inbetween training and classification phases

Note: This function must be called inbetween the training and classification phases

Parameters

doClean Flag indicating if the memory, keeping the trining data should be released after training

Implements DirectGraphicalModels::ITrain.

Reimplemented in DirectGraphicalModels::CTrainLinkNested< Trainer >.

Definition at line 60 of file TrainLink.h.

Member Data Documentation

◆ m_nStatesBase

byte DirectGraphicalModels::CTrainLink::m_nStatesBase

protected

Number of states (classes) at the base layer of ML-CRF.

Definition at line 87 of file TrainLink.h.

◆ m_nStatesOccl

byte DirectGraphicalModels::CTrainLink::m_nStatesOccl

protected

Number of states (classes) at the occlusion layerts of ML-CRF.

Definition at line 88 of file TrainLink.h.

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

D:/Projects/DGM/modules/DGM/TrainLink.h
D:/Projects/DGM/modules/DGM/TrainLink.cpp

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ CTrainLink()

◆ ~CTrainLink()

Member Function Documentation

◆ addFeatureVec() [1/3]

◆ addFeatureVec() [2/3]

◆ addFeatureVec() [3/3]

◆ calculateLinkPotentials()

◆ getLinkPotentials()

◆ train()

Member Data Documentation

◆ m_nStatesBase

◆ m_nStatesOccl