#include <ImFusion/ML/KeypointExtractionAlgorithm.h>

Algorithm for extracting keypoints from blobs representing the probability distribution of the keypoint location. More...

Inheritance diagram for KeypointExtractionAlgorithm:

Detailed Description

Algorithm for extracting keypoints from blobs representing the probability distribution of the keypoint location.

Public Types
enum class	ExtractionMode { MAX = 0 , MEAN = 1 , LOCAL_MAX = 2 }
	specify how to aggregate blobs into keypoints Max: global max for each channel (1 point/channels) Mean: global mean for each channel (1 point/channels) Local Max: clustering of local blob maxima for each channel (multiple points/channels)

enum class	ClusterMergingMode { AVERAGE = 0 , MAX = 1 }
	In case the Local Max extraction mode is used, define how to merge clusters that are below the p_minClusterDist threshold. More...

Public Types inherited from Algorithm
enum	Status { Unknown = -1 , Success = 0 , Error = 1 , InvalidInput , IncompleteInput , OutOfMemoryHost , OutOfMemoryGPU , UnsupportedGPU , UnknownAction , AbortedByUser , User = 1000 }
	Status codes. More...

Public Member Functions
	KeypointExtractionAlgorithm (const SharedImageSet *image=nullptr)
	Constructor.

	~KeypointExtractionAlgorithm () override
	Destructor.

void	setImage (const SharedImageSet *image)
	image setter

void	compute () override
	Execute the algorithm.

void	configuration (Properties *p) const override
	Retrieve the properties of this object.

OwningDataList	takeOutput () override
	Generated output is added to the data list.

std::vector< std::vector< vec3 > >	extractKeypoints (const SharedImage &blobs) const
	Given the predicted blobs, aggregates them into a collection of points.

Public Member Functions inherited from Algorithm
	Algorithm ()
	Default constructor will registers a single "compute" action that calls compute() and returns status().

virtual void	setProgress (Progress *progress)
	Sets a Progress interface the algorithm can use to notify observers about its computing progress.

Progress *	progress () const
	Returns the progress interface if set.

virtual int	status () const
	Indicates the status of the last call to compute().

virtual bool	survivesDataDeletion (const Data *) const
	Indicates whether the algorithm can handle (partial) deletion of the specified data, by default this checks whether the data is in the input list.

const FactoryInfo &	factoryInfo () const
	Returns the record describing how this Algorithm was instantiated by the AlgorithmFactory.

void	setFactoryInfo (const FactoryInfo &value)
	Sets the record describing how this Algorithm was instantiated by the AlgorithmFactory.

Status	runAction (const std::string &id)
	Run the action with name `id` if it exists.

const std::vector< Action > &	actions ()
	Get a mapping from Action id to Action as registered in this algorithm.

Public Member Functions inherited from Configurable
virtual void	configure (const Properties *p)
	Configure this object instance by de-serializing the given Properties.

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

Static Public Member Functions
static bool	createCompatible (const DataList &data, Algorithm **a=nullptr)
	If data is compatible with algorithm return true. If a is not zero, create algorithm with input data.

static std::vector< std::tuple< vec3, double > >	clusterSingleChannelBlob (const SharedImageSet *blob, double blobIntensityCutoff, double minClusterDistance, double minClusterWeight, size_t maxInternalClusters=1000, bool runSmoothing=false, int smoothingHalfKernel=2, bool intensityBasedRefinement=false, ClusterMergingMode mergineMode=ClusterMergingMode::AVERAGE)
	Aggregates the cluster found in single blob channel into different cluster centers representing different instances of the same landmark class.

static void	runKMeans (std::vector< vec3 > &inOutCenters, std::vector< double > &outCenterWeights, const std::vector< vec3 > &voxels, const std::vector< double > &voxelWeights, std::vector< mat3 > &inOutCovs, std::vector< int > &outVoxelMembership)
	K-mean clustering.

static std::vector< vec3 >	initKMeansPlusPlus (const std::vector< vec3 > &voxels, int numClusters, Random::Generator &generator=Random::globalGenerator())
	Compute cluster center initialization with K-means++.

Static Public Member Functions inherited from Algorithm
static bool	createCompatible (const DataList &data, Algorithm **a=nullptr)
	Factory function to check algorithm compatibility with input data and optionally instantiate it.

Public Attributes
Parameter< ExtractionMode >	p_extractionMode = {"keypointExtractionMode", ExtractionMode::MAX, this}
	How to aggregate the blobs into points See Documentation for ExtractionMode enum.

Parameter< double >	p_blobIntensityCutoff = {"blobIntensityCutoff", 0.02, this}
	Minimum blob intensity to be considered in analysis.

Parameter< double >	p_minClusterDist = {"minClusterDistance", 10., this}
	In case of local aggregation methods, minimum distance allowed among clusters.

Parameter< double >	p_minClusterWeight = {"minClusterWeight", 0.1, this}
	In case of local aggregation methods, minimum intensity for cluster to be consider independent.

Parameter< ClusterMergingMode >	p_clusterMergingMode = {"Cluster Merging Mode", ClusterMergingMode::AVERAGE, this}
	In case of local aggregation methods, minimum distance allowed among clusters.

Parameter< size_t >	p_maxInternalClusters = {"maxInternalClusters", 1000, this}
	In case of local aggregation methods, maximum number of internal clusters to be considered; to avoid excessive numbers that stall the algorithm.

Parameter< bool >	p_runSmoothing = {"runSmoothing", false, this}
	Runs a Gaussian smoothing with 1 pixel standard deviation to improve stability of local maxima.

Parameter< int >	p_smoothingHalfKernel = {"smoothingHalfKernel", 2, this}
	Runs a Gaussian smoothing with 1 pixel standard deviation to improve stability of local maxima.

Parameter< bool >	p_intensityBasedRefinement = {"runIntensityRefinement", false, this}
	Runs blob intensity based refinement of clustered keypoints.

Public Attributes inherited from Algorithm
Signal	signalOutputChanged
	Signal should be emitted by Algorithms when their output/result has changed.

Signal	signalParametersChanged
	Signal should be emitted by Algorithms when their parameter configuration has changed.

Public Attributes inherited from Configurable
Signal	signalParametersChanged
	Emitted whenever one of the registered Parameters' or SubPropertys' `signalValueChanged` signal was emitted.

Additional Inherited Members
Protected Member Functions inherited from Algorithm
void	loadDefaults ()

void	registerAction (const std::string &id, const std::string &guiName, const std::function< Algorithm::Status(void)> &action)
	Register an action to be run via runAction.

template<typename D>
void	registerAction (const std::string &id, const std::string &guiName, Algorithm::Status(D::*action)(void))
	Template version of runAction that can be used with a pointer to a member function.

void	registerAction (const Action &action)
	Register an action.

Protected Attributes inherited from Algorithm
std::string	m_name
	Algorithm name.

Progress *	m_progress = nullptr
	Non-owing pointer to a progress interface. May be a nullptr.

FactoryInfo	m_factoryInfo = {}
	Record describing how this algorithm was instantiated by the AlgorithmFactory.

int	m_status = Status::Unknown
	Algorithm status after last call to compute()

std::vector< Action >	m_actions
	Map of key given by the id of the action, of the available actions of this algorithm.

Protected Attributes inherited from Configurable
std::vector< Param >	m_params
	List of all registered Parameter and SubProperty instances.

Member Enumeration Documentation

◆ ClusterMergingMode

enum class ClusterMergingMode

strong

In case the Local Max extraction mode is used, define how to merge clusters that are below the p_minClusterDist threshold.

If AVERAGE, neighbouring clusters below the threshold are averaged. If MAX, the cluster with the highest intensity is kept and the other removed.

Member Function Documentation

◆ compute()

void compute ( )

overridevirtual

Execute the algorithm.

Implements Algorithm.

◆ configuration()

void configuration ( Properties * p ) const

overridevirtual

Retrieve the properties of this object.

Reimplemented from Configurable.

◆ takeOutput()

OwningDataList takeOutput ( )

overridevirtual

Generated output is added to the data list.

Reimplemented from Algorithm.

◆ extractKeypoints()

std::vector< std::vector< vec3 > > extractKeypoints ( const SharedImage & blobs ) const

Given the predicted blobs, aggregates them into a collection of points.

Specifically, the outer vector lists the different landmark classes (number of blob channels), while the inner vector lists the different instances of the same landmark class found in a single blob channel.

◆ clusterSingleChannelBlob()

static std::vector< std::tuple< vec3, double > > clusterSingleChannelBlob	(	const SharedImageSet *	blob,
		double	blobIntensityCutoff,
		double	minClusterDistance,
		double	minClusterWeight,
		size_t	maxInternalClusters = 1000,
		bool	runSmoothing = false,
		int	smoothingHalfKernel = 2,
		bool	intensityBasedRefinement = false,
		ClusterMergingMode	mergineMode = ClusterMergingMode::AVERAGE )

static

Aggregates the cluster found in single blob channel into different cluster centers representing different instances of the same landmark class.

Each cluster center is bundled with its own cluster weight (vec3, double). This function is used in cases where local filtering of the blob is required, i.e. for multi-instance landmark detection. Use this function if you need to further manipulate the aggregated clusters.

◆ runKMeans()

static void runKMeans	(	std::vector< vec3 > &	inOutCenters,
		std::vector< double > &	outCenterWeights,
		const std::vector< vec3 > &	voxels,
		const std::vector< double > &	voxelWeights,
		std::vector< mat3 > &	inOutCovs,
		std::vector< int > &	outVoxelMembership )

static

K-mean clustering.

Inputs are the centers, voxels, and weights for each voxel. Outputs are the new centers, center weights, covariances of each center, and new membership of each voxel.

◆ initKMeansPlusPlus()

static std::vector< vec3 > initKMeansPlusPlus	(	const std::vector< vec3 > &	voxels,
		int	numClusters,
		Random::Generator &	generator = Random::globalGenerator() )

static

Compute cluster center initialization with K-means++.

Non-deterministic algorithm, so the result may vary between runs.

Note: Uses the optional Random::Generator for random number generation, Random::globalGenerator by default. This means one can pass a specific generator to ensure reproducibility.

Parameters

voxels	The voxels to cluster
numClusters	The number of clusters to initialize

Returns: The cluster centers

Member Data Documentation

◆ p_maxInternalClusters

Parameter<size_t> p_maxInternalClusters = {"maxInternalClusters", 1000, this}

In case of local aggregation methods, maximum number of internal clusters to be considered; to avoid excessive numbers that stall the algorithm.

If there are more, the lower weighted ones are removed first.

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

ImFusion/ML/KeypointExtractionAlgorithm.h

Detailed Description

Public Types

Public Member Functions

Static Public Member Functions

Public Attributes

Additional Inherited Members

Member Enumeration Documentation

◆ ClusterMergingMode

Member Function Documentation

◆ compute()

◆ configuration()

◆ takeOutput()

◆ extractKeypoints()

◆ clusterSingleChannelBlob()

◆ runKMeans()

◆ initKMeansPlusPlus()

Member Data Documentation

◆ p_maxInternalClusters