Package deepnetts.data

Class ImageSet

java.lang.Object

javax.visrec.ml.data.BasicDataSet<ExampleImage>

deepnetts.data.TabularDataSet<ExampleImage>

deepnetts.data.ImageSet

All Implemented Interfaces:

Iterable<ExampleImage>, javax.visrec.ml.data.DataSet<ExampleImage>

public class ImageSet
extends TabularDataSet<ExampleImage>

Data set with images that will be used to train convolutional neural network.

Author:

Zoran Sevarac

Nested Class Summary

Nested classes/interfaces inherited from class deepnetts.data.TabularDataSet

TabularDataSet.Item

Constructor Summary

Constructors

Constructor	Description
ImageSet(int imageWidth, int imageHeight)	Creates empty image set for images of specified size.
ImageSet(int imageWidth, int imageHeight, String imageDirPath)	Creates image set with images from specified directory path.

Method Summary

Modifier and Type	Method	Description
javax.visrec.ml.data.DataSet<ExampleImage>	add(ExampleImage exImage)	Adds an example image that will be used to train deep learning model to this set.
Map<String,Integer>	countByClasses()
boolean	getBrightness()
boolean	getCropCornersAndCenter()
String	getDelimiter()
boolean	getFlipHorizontal()
boolean	getGrayscale()
int	getImageHeight()
int	getImageWidth()
boolean	getInvertImages()
int	getLabelsCount()
TensorBase	getMean()
ImageResize	getResizeStrategy()
boolean	getScaleImages()	Returns flag that indicates wheather images should be scaled to specified dimensions while creating image set.
String[]	getTargetColumnsNames()	Returns output/image labels.
boolean	getTranslate()
boolean	getZeroMeanPixels()
void	loadImages(File imageIdxFile)	Loads example images with corresponding labels from the specified file.
void	loadImages(File imageIdxFile, int numOfImages)	Loads specified number of example images with corresponding labels from the specified file.
void	loadImages(String imageIdxFile)	Loads images from the specified image index file.
String[]	loadLabels(File file)	Loads and returns image labels to train neural network from the specified file.These labels will be used to label network's outputs.
String[]	loadLabels(String filePath)	Loads and returns image labels to train neural network from the specified file.
void	setBrightness(boolean brightness)
void	setCropCornersAndCenter(boolean cropCornersAndCenter)
void	setDelimiter(String delimiter)
void	setFlipHorizontal(boolean flipHorizontal)
void	setGrayscale(boolean grayscale)
ImageSet	setInvertImages(boolean invertImages)
ImageSet	setResizeStrategy(ImageResize resizeStrategy)
final void	setScaleImages(boolean scaleImages)
void	setTranslate(boolean translate)
ImageSet[]	split(double... partSizes)	Splits data set into several parts specified by the input parameter partSizes.
TensorBase	zeroMean()	Applies zero mean normalization to entire dataset, and returns mean tensor.
TensorBase	zeroMeanAndNormalize()
TensorBase	zeroMeanPerChannel()

Methods inherited from class deepnetts.data.TabularDataSet

countMissingValues, countMissingValues, getColumnNames, getNumInputs, getNumOutputs, hasMissingValues, hasMissingValues, setColumnNames, shuffle, shuffle, split, trainTestSplit

Methods inherited from class javax.visrec.ml.data.BasicDataSet

getColumns, getItems, setAsTargetColumns, setAsTargetColumns, setColumns

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface javax.visrec.ml.data.DataSet

addAll, clear, get, isEmpty, iterator, shuffle, size, split, split, split, stream

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Constructor Details

ImageSet

public ImageSet(int imageWidth,
 int imageHeight)

Creates empty image set for images of specified size.

Parameters:

imageWidth -

imageHeight -

ImageSet

public ImageSet(int imageWidth,
 int imageHeight,
 String imageDirPath)
         throws IOException

Creates image set with images from specified directory path.

Parameters:

imageWidth - all images in set will be scaled to this width

imageHeight - all images in set will be scaled to this height

imageDirPath - path to the root directory which contains subdirectories with images

Throws:

IOException

Method Details

add

public javax.visrec.ml.data.DataSet<ExampleImage> add(ExampleImage exImage)
                                               throws DeepNettsException

Adds an example image that will be used to train deep learning model to this set.

Parameters:

exImage - holds raw image pixel data and corresponding image label

Returns:

this instance of ImageSet

Throws:

DeepNettsException - if image is empty or has wrong dimensions.

loadImages

public void loadImages(String imageIdxFile)
                throws FileNotFoundException

Loads images from the specified image index file.

Parameters:

imageIdxFile - txt file with list of images

Throws:

FileNotFoundException

loadImages

public void loadImages(File imageIdxFile)
                throws FileNotFoundException

Loads example images with corresponding labels from the specified file.

Parameters:

imageIdxFile - Plain text file that contains space delimited image paths and labels

Throws:

FileNotFoundException - if imageIdxFile was not found

loadImages

public void loadImages(File imageIdxFile,
 int numOfImages)
                throws DeepNettsException

Loads specified number of example images with corresponding labels from the specified file.

Parameters:

imageIdxFile - Plain text file which contains space delimited image file paths and label

numOfImages - number of images to load

Throws:

DeepNettsException

getLabelsCount

public int getLabelsCount()

split

public ImageSet[] split(double... partSizes)

Splits data set into several parts specified by the input parameter partSizes. Values of partSizes parameter represent the sizes of data set parts that will be returned. Part sizes are integer values that represent percents, cannot be negative or zero, and their sum must be 100

Specified by:

split in interface javax.visrec.ml.data.DataSet<ExampleImage>

Overrides:

split in class TabularDataSet<ExampleImage>

Parameters:

partSizes - sizes of the parts in percents

Returns:

parts of the data set of specified size

loadLabels

public String[] loadLabels(String filePath)
                    throws DeepNettsException

Loads and returns image labels to train neural network from the specified file. These labels will be used to label network's outputs.

Parameters:

filePath -

Returns:

Throws:

DeepNettsException

loadLabels

public String[] loadLabels(File file)
                    throws DeepNettsException

Loads and returns image labels to train neural network from the specified file.These labels will be used to label network's outputs.

Parameters:

file - file to load labels from

Returns:

Throws:

DeepNettsException

zeroMean

public TensorBase zeroMean()

Applies zero mean normalization to entire dataset, and returns mean tensor.

Returns:

mean Tensor for the entire dataset

zeroMeanPerChannel

public TensorBase zeroMeanPerChannel()

zeroMeanAndNormalize

public TensorBase zeroMeanAndNormalize()

getScaleImages

public boolean getScaleImages()

Returns flag that indicates wheather images should be scaled to specified dimensions while creating image set.

Returns:

setScaleImages

public final void setScaleImages(boolean scaleImages)

getInvertImages

public boolean getInvertImages()

setInvertImages

public ImageSet setInvertImages(boolean invertImages)

getFlipHorizontal

public boolean getFlipHorizontal()

getBrightness

public boolean getBrightness()

getGrayscale

public boolean getGrayscale()

setFlipHorizontal

public void setFlipHorizontal(boolean flipHorizontal)

setBrightness

public void setBrightness(boolean brightness)

setGrayscale

public void setGrayscale(boolean grayscale)

getTranslate

public boolean getTranslate()

setTranslate

public void setTranslate(boolean translate)

getCropCornersAndCenter

public boolean getCropCornersAndCenter()

setCropCornersAndCenter

public void setCropCornersAndCenter(boolean cropCornersAndCenter)

getTargetColumnsNames

public String[] getTargetColumnsNames()

Returns output/image labels.

Specified by:

getTargetColumnsNames in interface javax.visrec.ml.data.DataSet<ExampleImage>

Overrides:

getTargetColumnsNames in class TabularDataSet<ExampleImage>

Returns:

countByClasses

public Map<String,Integer> countByClasses()

getDelimiter

public String getDelimiter()

setDelimiter

public void setDelimiter(String delimiter)

getResizeStrategy

public ImageResize getResizeStrategy()

setResizeStrategy

public ImageSet setResizeStrategy(ImageResize resizeStrategy)

getImageWidth

public int getImageWidth()

getImageHeight

public int getImageHeight()

getMean

public TensorBase getMean()

getZeroMeanPixels

public boolean getZeroMeanPixels()