An EDataAugmentation object is responsible for storing the data augmentation parameters and generating new transformed images from existing ones. Deep learning algorithms are not invariant to the location, scale, or rotation of the elements of interest in the image. Thus, if the final application requires the deep learning algorithm to be invariant to those characteristics, the dataset must contain images covering the spectrum of those characteristics. More...

Inheritance diagram for eVision.EasyDeepLearning.EDataAugmentation:

[legend]

Collaboration diagram for eVision.EasyDeepLearning.EDataAugmentation:

[legend]

Public Member Functions
	EDataAugmentation ()
	Creates an EDataAugmentation object.
	EDataAugmentation (eVision.EasyDeepLearning.EDataAugmentation other)
	Creates a copy of the EDataAugmentation object other.
void	CopyTo (eVision.EasyDeepLearning.EDataAugmentation other)
	Copies itself to the EDataAugmentation object other.
void	Load (string path)
	Loads an EDataAugmentation object. The given ESerializer must have been created for reading.
void	Load (eVision.ESerializer serializer)
	Loads an EDataAugmentation object. The given ESerializer must have been created for reading.
void	Save (string path)
	Saves an EDataAugmentation object. The given ESerializer must have been created for writing.
void	Save (eVision.ESerializer serializer)
	Saves an EDataAugmentation object. The given ESerializer must have been created for writing.
bool	HasDataAugmentations ()
	Whether the EDataAugmentation object contains augmentations.
eVision.EBaseROI	Generate (eVision.EBaseROI img, eVision.EasyDeepLearning.EDLDataAugmentationType type)
	Generates a new image from a generic EBaseROI image. The caller is responsible for calling freeing the memory of the returned image by calling delete on the image.
eVision.EBaseROI	Generate (eVision.EBaseROI img)
	Generates a new image from a generic EBaseROI image. The caller is responsible for calling freeing the memory of the returned image by calling delete on the image.
void	Generate (eVision.EROIBW8 source, eVision.EROIBW8 dest, eVision.EasyDeepLearning.EDLDataAugmentationType type)
	Generates a new image.
void	Generate (eVision.EROIBW8 source, eVision.EROIBW8 dest)
	Generates a new image.
void	Generate (eVision.EROIBW16 source, eVision.EROIBW16 dest, eVision.EasyDeepLearning.EDLDataAugmentationType type)
	Generates a new image.
void	Generate (eVision.EROIBW16 source, eVision.EROIBW16 dest)
	Generates a new image.
void	Generate (eVision.EROIC24 source, eVision.EROIC24 dest, eVision.EasyDeepLearning.EDLDataAugmentationType type)
	Generates a new image.
void	Generate (eVision.EROIC24 source, eVision.EROIC24 dest)
	Generates a new image.

Properties
int	MaxHorizontalShift `[get, set]`
	Return the maximum horizontal shift allowed.
int	MaxVerticalShift `[get, set]`
	Return the maximum vertical shift allowed.
float	MaxRotationAngle `[get, set]`
	Returns the maximum rotation angle allowed. The unit of the returned angle is in the current angle unit (see Easy.AngleUnit).
bool	EnableVerticalFlip `[get, set]`
	Gets whether to use vertically flipped versions of input images.
bool	EnableHorizontalFlip `[get, set]`
	Gets whether to use horizontally flipped versions of input images.
float	MinScale `[get, set]`
	Gets the minimum scaling allowed.
float	MaxScale `[get, set]`
	Gets the maximum scaling allowed.
float	MaxHorizontalShear `[get, set]`
	Gets the maximum absolute horizontal shear, represented as an angle from the vertical direction.
float	MaxVerticalShear `[get, set]`
	Gets the maximum absolute vertical shear, represented as an angle from the horizontal direction.
float	MaxBrightnessOffset `[get, set]`
	Gets the maximum absolute brightness offset.
float	MaxContrastGain `[get, set]`
	Gets the maximum contrast gain.
float	MinContrastGain `[get, set]`
	Gets the minimum contrast gain.
float	MinGamma `[get, set]`
	Gets the minimum gamma for gamma correction.
float	MaxGamma `[get, set]`
	Gets the maximum gamma for gamma correction.
float	MaxHueOffset `[get, set]`
	Gets the maximum absolute hue offset.
float	MinSaturationGain `[get, set]`
	Gets the minimum saturation gain.
float	MaxSaturationGain `[get, set]`
	Gets the maximum saturation gain.
float	GaussianNoiseMaximumStandardDeviation `[get, set]`
	The Gaussian noise maximum standard deviation.
float	GaussianNoiseMinimumStandardDeviation `[get, set]`
	The Gaussian noise minimum standard deviation.
float	SpeckleNoiseMaximumStandardDeviation `[get, set]`
	The speckle noise maximum standard deviation.
float	SpeckleNoiseMinimumStandardDeviation `[get, set]`
	The speckle noise minimum standard deviation.
float	SaltAndPepperNoiseMaximumDensity `[get, set]`
	The maximum density of the salt and pepper noise.
float	SaltAndPepperNoiseMinimumDensity `[get, set]`
	The minimum density of the salt and pepper noise.
uint	StainDisruptionMaxAnchorPoints `[get, set]`
	The maximum number of anchor points to use while disrupting the ellipse.
float	StainDisruptionMaxIntensity `[get, set]`
	The maximum offset applied to the coordinates (x,y) of a pixel from the ellipse contour.
uint	StainEllipseMinRadius `[get, set]`
	The minimum radius of the ellipse used to produce the stain.
uint	StainEllipseMaxRadius `[get, set]`
	The maximum radius of the ellipse used to produce the stain.
byte	MinStainColor `[get, set]`
	The minimum color value from which to draw a color to fill in the stain (we use a gaussian distribution of with a mean between EDataAugmentation.MinStainColor and EDataAugmentation.MaxStainColor).
byte	MaxStainColor `[get, set]`
	The maximum color value from which to draw a color to fill in the stain (we use a gaussian distribution of with a mean between EDataAugmentation.MinStainColor and EDataAugmentation.MaxStainColor).
uint	StainColorVariation `[get, set]`
	The variation of the color used to fill in the ellipse (We use an gaussian distribution of standard mean deviation stainColorVariation).
float	StainInterpolationSiteFactor `[get, set]`
	The factor to compute the number of interpolation sites to use from the number of anchor points.
float	StainProbability `[get, set]`
	Probability of applying the stain data augmentation (default value: 0, i.e. no stain).
int	StainBlur `[get, set]`
	Stain blur.

Detailed Description

An EDataAugmentation object is responsible for storing the data augmentation parameters and generating new transformed images from existing ones. Deep learning algorithms are not invariant to the location, scale, or rotation of the elements of interest in the image. Thus, if the final application requires the deep learning algorithm to be invariant to those characteristics, the dataset must contain images covering the spectrum of those characteristics.

Data augmentation can be used to avoid capturing in the original dataset the whole spectrum of those characteristics by automatically generating new versions of the images in the dataset that are shifted, scaled or rotated to cover this spectrum.

When generating a new image, the EDataAugmentation will randomly pick transformation value within the specified limits.

When using data augmentation, one must carefully chose the spectrum of transformations such that the element of interest are still visible in the image.

License(s): EasyClassify or EasySegment or EasyLocate

Constructor & Destructor Documentation

◆ EDataAugmentation() [1/2]

eVision.EasyDeepLearning.EDataAugmentation.EDataAugmentation ( )

Creates an EDataAugmentation object.

◆ EDataAugmentation() [2/2]

eVision.EasyDeepLearning.EDataAugmentation.EDataAugmentation ( eVision.EasyDeepLearning.EDataAugmentation other )

Creates a copy of the EDataAugmentation object other.

Parameters

other The other EDataAugmentation that we will copy into this one.

Member Function Documentation

◆ CopyTo()

void eVision.EasyDeepLearning.EDataAugmentation.CopyTo ( eVision.EasyDeepLearning.EDataAugmentation other )

Copies itself to the EDataAugmentation object other.

Parameters

other The other EDataAugmentation that we will copy into this one.

◆ Generate() [1/8]

eVision.EBaseROI eVision.EasyDeepLearning.EDataAugmentation.Generate ( eVision.EBaseROI img )

Generates a new image from a generic EBaseROI image. The caller is responsible for calling freeing the memory of the returned image by calling delete on the image.

Parameters

img	The image to transform.