ClassifierFactory Class Reference

Learner object that creates a classifier from an image list. More...

Classes
class	LearnProgressEventArgs
	Arguments for the LearnProgress event. More...

Public Member Functions
	ClassifierFactory ()
	Construct a learner and initialize it with an image list.
Classifier	Learn (SampleImageList sampleImageList, out double learnResult)
	Learn a new classifier from a SampleImageList using the currently set parameters. More...

Public Attributes
readonly string	PreprocessingCodeDefault = string.Empty
	Default preprocessing code used by the learner.
DecisionBoundary	BOUNDARYDEFAULT = DecisionBoundary.Soft
	Default decision boundary the leaner will use.

Static Public Attributes
const InstanceGeneration	InstanceGenerationDefault = InstanceGeneration.Random
	The default type of instance generation that the learner uses.
const int	SamplesPerImageDefault = 1
	The default number of samples per image the learner uses.
const Symmetries	SymmetriesDefault = Symmetries.Identity
	Default setting for Symmetries that the learner uses.
const int	RotationRangeDefault = 0
	Default rotation range used by the learner.
static readonly Point2D	TranslationRangeDefault = new Point2D(0, 0)
	Default translation range used by the learner.

Properties
InstanceGeneration	InstanceGeneration [get, set]
	Defines/gets the approach to be taken to additional instance generation during the learning phase.
int	SamplesPerImage [get, set]
	Defines how many samples should be used per sample inside the SampleImageList during classifier generation. Setting this property to a value > 1 will lead to additional temporary images being generated during the learning phase based on the settings of the properties InstanceGeneration, Symmetries, RotationRange and TranslationRange. More...
Symmetries	Symmetries [get, set]
	Gets or sets whether the images from the SampleImageList property should be mirror on the X axis during learning. More...
int	RotationRange [get, set]
	Gets/Sets the limit of the range of rotation to be used for automatic temporary instance generation during the learning phase. Rotations may range from -RotationRange to +RotationRange during the learning phase. Note that the actual number of rotations generated during the learning phase depends on the setting of the SamplesPerImage property. RotationRange is measured in degrees. If the rotation range set through this property is bigger than what the SIL provided for learning allows, the maximum allowable rotation will be used during learning. More...
Point2D	TranslationRange [get, set]
	Gets/Sets the limit of the range of translation to be used for automatic temporary instance generation during the learning phase. Translations may range from -TranslationRange to +TranslationRange during the learning phase. Note that the actual number of translations generated during the learning phase depends on the setting of the SamplesPerImage property. TranslationRange is measured in pixels. If the translation range set through this property is bigger than what the SIL provided for learning allows, the maximum allowable translation will be used during learning. More...
string	PreprocessingCode [get, set]
	Preprocessing code to be used for the learning phase; valid preprocessing codes consist of characters 'a' and 'p' (or an empty string); the length of the preprocessing code should be compatible with the image list's feature window geometry (this will be checked when the Learn method is invoked). More...
DecisionBoundary	Boundary [get, set]
	Decision boundary setting to be used for learning the classifier.

Events
EventHandler< LearnProgressEventArgs >	LearnProgress
	Event that will inform about the progress of an ongoing learn operation.

Detailed Description

Learner object that creates a classifier from an image list.

Member Function Documentation

Learn()

Classifier Learn	(	SampleImageList	sampleImageList,
		out double	learnResult
	)

Learn a new classifier from a SampleImageList using the currently set parameters.

Parameters

sampleImageList

image list from which to learn the classifier

learnResult

the learn result; if learning was completed successfully this parameter will contain the learn result which is a measure for the amount of decision boundary_ violation that occurred during the learn process. This value can be interpreted as a measure for the separability of the image list from which the classifier was learned under the parameter conditions with which is was learned; it will not necessarily be a measure for the actual quality of the resulting classifier.

Returns

A new Manto classifier generated based on the ClassifierFactory's settings and the training data provided in sampleImageList .

Exceptions

ObjectDisposedException	If the SampleImageList or the learner has been disposed
ArgumentNullException	when the sampleImageList parameter is null
ArgumentException	If the currently set PreprocessingCode is not suitable for the feature window size of the sampleImageList

Property Documentation

PreprocessingCode

string PreprocessingCode

getset

Preprocessing code to be used for the learning phase; valid preprocessing codes consist of characters 'a' and 'p' (or an empty string); the length of the preprocessing code should be compatible with the image list's feature window geometry (this will be checked when the Learn method is invoked).

Exceptions

ArgumentException	when the preprocessing code contains invalid characters
ArgumentNullException	when trying to set a null value as the preprocessing code

RotationRange

int RotationRange

getset

Gets/Sets the limit of the range of rotation to be used for automatic temporary instance generation during the learning phase. Rotations may range from -RotationRange to +RotationRange during the learning phase. Note that the actual number of rotations generated during the learning phase depends on the setting of the SamplesPerImage property. RotationRange is measured in degrees. If the rotation range set through this property is bigger than what the SIL provided for learning allows, the maximum allowable rotation will be used during learning.

Exceptions

ArgumentException

when trying to set a value less than 0 or bigger than 180. SampleImageList

SamplesPerImage

int SamplesPerImage

getset

Defines how many samples should be used per sample inside the SampleImageList during classifier generation. Setting this property to a value > 1 will lead to additional temporary images being generated during the learning phase based on the settings of the properties InstanceGeneration, Symmetries, RotationRange and TranslationRange.

Note that for each additional image beyond 1 being used, the amount of memory required during the learning learnStage will effectively be incremented by the the size of one Image List. Or, to put it differently, if 5 samples per image are being used, Manto will need five times the amount of memory (and time) it would take to learn the classifier compared to the use of only 1 sample per image. This consumption multiplies with the requirement generated by the setting of the Symmetries property. SamplesPerImage and Symmetries should therefore be used with caution as you can easily generate a situation where the classifier can no longer be generated from the Image List because Manto runs out of memory during the learning phase.

Exceptions

ArgumentException

when trying to set a value less than 1

Symmetries

Symmetries Symmetries

getset

Gets or sets whether the images from the SampleImageList property should be mirror on the X axis during learning.

Exceptions

ArgumentException

when trying to set the value Symmetries.None

Note that for each symmetry being used, the amount of memory required during the learning learnStage will effectively be incremented by the the size of one Image List. Or, to put it differently, if all symmetries are being used, Manto will need four times the amount of memory (and time) it would take to learn the classifier compared to the size of the image list. This consumption multiplies with the requirement generated by the setting of the SamplesPerImage property. SamplesPerImage and Symmetries should therefore be used with caution as you can easily generate a situation where the classifier can no longer be generated from the Image List because Manto runs out of memory during the learning phase.

TranslationRange

Point2D TranslationRange

getset

Gets/Sets the limit of the range of translation to be used for automatic temporary instance generation during the learning phase. Translations may range from -TranslationRange to +TranslationRange during the learning phase. Note that the actual number of translations generated during the learning phase depends on the setting of the SamplesPerImage property. TranslationRange is measured in pixels. If the translation range set through this property is bigger than what the SIL provided for learning allows, the maximum allowable translation will be used during learning.

Exceptions

ArgumentException

when trying to set a value less than 0 SampleImageList