curfil::ImageFeatureEvaluation Class Reference

Helper class for the four phases in the cost-intensive best-split evaluation during random forest training. More...

#include <random_tree_image.h>

Public Member Functions
	ImageFeatureEvaluation (const size_t treeId, const TrainingConfiguration &configuration)
	helper object for feature evaluation for the given tree and training configuration
std::vector< SplitFunction < PixelInstance, ImageFeatureFunction > >	evaluateBestSplits (RandomSource &randomSource, const std::vector< std::pair< boost::shared_ptr< RandomTree< PixelInstance, ImageFeatureFunction > >, std::vector< const PixelInstance * > > > &samplesPerNode)
std::vector< std::vector < const PixelInstance * > >	prepare (const std::vector< const PixelInstance * > &samples, RandomTree< PixelInstance, ImageFeatureFunction > &node, cuv::host_memory_space)
std::vector< std::vector < const PixelInstance * > >	prepare (const std::vector< const PixelInstance * > &samples, RandomTree< PixelInstance, ImageFeatureFunction > &node, cuv::dev_memory_space, bool keepMutexLocked=true)
ImageFeaturesAndThresholds < cuv::host_memory_space >	generateRandomFeatures (const std::vector< const PixelInstance * > &batches, int seed, const bool sort, cuv::host_memory_space)
ImageFeaturesAndThresholds < cuv::dev_memory_space >	generateRandomFeatures (const std::vector< const PixelInstance * > &batches, int seed, const bool sort, cuv::dev_memory_space)
template<class memory_space >
void	sortFeatures (ImageFeaturesAndThresholds< memory_space > &featuresAndThresholds, const cuv::ndarray< int, memory_space > &keysIndices) const
	sort features by the given keys
template<class memory_space >
cuv::ndarray< WeightType, memory_space >	calculateFeatureResponsesAndHistograms (RandomTree< PixelInstance, ImageFeatureFunction > &node, const std::vector< std::vector< const PixelInstance * > > &batches, const ImageFeaturesAndThresholds< memory_space > &featuresAndThresholds, cuv::ndarray< FeatureResponseType, cuv::host_memory_space > *featureResponsesHost=0)
template<class memory_space >
cuv::ndarray< ScoreType, cuv::host_memory_space >	calculateScores (const cuv::ndarray< WeightType, memory_space > &counters, const ImageFeaturesAndThresholds< memory_space > &featuresAndThresholds, const cuv::ndarray< WeightType, memory_space > &histogram)

Detailed Description

Helper class for the four phases in the cost-intensive best-split evaluation during random forest training.

See the Master’s thesis "Accelerating Random Forests on CPUs and GPUs for Object-Class Image Segmentation" for more details on this implementation.

Clients are not intended to use this class directly.

Definition at line 961 of file random_tree_image.h.

Member Function Documentation

template<class memory_space >

cuv::ndarray<WeightType, memory_space> curfil::ImageFeatureEvaluation::calculateFeatureResponsesAndHistograms	(	RandomTree< PixelInstance, ImageFeatureFunction > &	node,
		const std::vector< std::vector< const PixelInstance * > > &	batches,
		const ImageFeaturesAndThresholds< memory_space > &	featuresAndThresholds,
		cuv::ndarray< FeatureResponseType, cuv::host_memory_space > *	featureResponsesHost = 0
	)

Returns

histogram counters after calculating feature responses

template<class memory_space >

cuv::ndarray<ScoreType, cuv::host_memory_space> curfil::ImageFeatureEvaluation::calculateScores	(	const cuv::ndarray< WeightType, memory_space > &	counters,
		const ImageFeaturesAndThresholds< memory_space > &	featuresAndThresholds,
		const cuv::ndarray< WeightType, memory_space > &	histogram
	)

Returns

normalized cores after calculating the information gain that results from splitting

std::vector<SplitFunction<PixelInstance, ImageFeatureFunction> > curfil::ImageFeatureEvaluation::evaluateBestSplits	(	RandomSource &	randomSource,
		const std::vector< std::pair< boost::shared_ptr< RandomTree< PixelInstance, ImageFeatureFunction > >, std::vector< const PixelInstance * > > > &	samplesPerNode
	)

Returns

best features and thresholds to split after evaluating splits

ImageFeaturesAndThresholds<cuv::host_memory_space> curfil::ImageFeatureEvaluation::generateRandomFeatures	(	const std::vector< const PixelInstance * > &	batches,
		int	seed,
		const bool	sort,
		cuv::host_memory_space
	)

Returns

random features and thresholds

ImageFeaturesAndThresholds<cuv::dev_memory_space> curfil::ImageFeatureEvaluation::generateRandomFeatures	(	const std::vector< const PixelInstance * > &	batches,
		int	seed,
		const bool	sort,
		cuv::dev_memory_space
	)

Returns

random features and thresholds

std::vector<std::vector<const PixelInstance*> > curfil::ImageFeatureEvaluation::prepare	(	const std::vector< const PixelInstance * > &	samples,
		RandomTree< PixelInstance, ImageFeatureFunction > &	node,
		cuv::host_memory_space
	)

Returns

a batch that contains the given samples

std::vector<std::vector<const PixelInstance*> > curfil::ImageFeatureEvaluation::prepare	(	const std::vector< const PixelInstance * > &	samples,
		RandomTree< PixelInstance, ImageFeatureFunction > &	node,
		cuv::dev_memory_space	,
		bool	keepMutexLocked = true
	)

Returns

a batch that contains the given samples

---

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

• src/curfil/random_tree_image.h