weka.dl4j.zoo

Class AbstractZooModel

java.lang.Object

weka.dl4j.zoo.AbstractZooModel

All Implemented Interfaces:

java.io.Serializable, weka.core.OptionHandler

Direct Known Subclasses:

CustomNet, Dl4jAlexNet, Dl4jDarknet19, Dl4jFaceNetNN4Small2, Dl4jInceptionResNetV1, Dl4jLeNet, Dl4jResNet50, Dl4jSqueezeNet, Dl4jVGG, Dl4jXception, KerasDenseNet, KerasEfficientNet, KerasInceptionV3, KerasNASNet, KerasResNet, KerasVGG, KerasXception

public abstract class AbstractZooModel
extends java.lang.Object
implements weka.core.OptionHandler, java.io.Serializable

This class contains the logic necessary to load the pretrained weights for a given zoo model It also handles the addition/removal of output layers to enable training the model in DL4J.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
AbstractZooModel()

Method Summary

Modifier and Type	Method and Description
boolean	getChannelsLast()
org.deeplearning4j.nn.graph.ComputationGraph	getDefaultGraph() Convenience method to returns a default pretrained graph for this zoo model
java.lang.String[]	getExtraLayersToRemove()
java.lang.String	getFeatureExtractionLayer()
abstract org.nd4j.linalg.dataset.api.preprocessor.ImagePreProcessingScaler	getImagePreprocessingScaler() Get the preprocessor to process this model's data with
abstract int[]	getInputShape() Get the input shape of this zoomodel
int	getNumFExtractOutputs()
java.lang.String[]	getOptions() Gets the current settings of the Classifier.
java.lang.String	getOutputlayer()
PretrainedType	getPretrainedType()
java.lang.String	getPrettyName()
abstract java.lang.Enum	getVariation() Get the current variation of the zoo model (e.g., Resnet50 or Resnet101)
abstract org.deeplearning4j.nn.graph.ComputationGraph	init(int numLabels, long seed, int[] shape, boolean filterMode) Initialize the ZooModel as MLP.
org.deeplearning4j.nn.graph.ComputationGraph	initZooModel(org.deeplearning4j.zoo.ZooModel zooModel, org.deeplearning4j.nn.graph.ComputationGraph defaultNet, long seed, int numLabels, boolean filterMode) Initialize the zoo model with the supplied params.
boolean	isPretrained()
boolean	isRequiresPooling()
java.util.Enumeration<weka.core.Option>	listOptions() Returns an enumeration describing the available options.
boolean	requiresPreProcessing() Does the model require input images to be preprocessed?.
void	setChannelsLast(boolean channelsLast)
void	setExtraLayersToRemove(java.lang.String[] m_extraLayersToRemove)
void	setFeatureExtractionLayer(java.lang.String m_featureExtractionLayer)
void	setNumFExtractOutputs(int m_numFExtractOutputs)
void	setOptions(java.lang.String[] options) Parses a given list of options.
void	setOutputLayer(java.lang.String m_outputLayer)
void	setPretrainedType(PretrainedType pretrainedType)
void	setRequiresPooling(boolean requiresPooling)

Methods inherited from class java.lang.Object

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

Methods inherited from interface weka.core.OptionHandler

makeCopy

Constructor Detail

AbstractZooModel

public AbstractZooModel()

Method Detail

init

public abstract org.deeplearning4j.nn.graph.ComputationGraph init(int numLabels,
                                                                  long seed,
                                                                  int[] shape,
                                                                  boolean filterMode)
                                                           throws java.lang.UnsupportedOperationException

Initialize the ZooModel as MLP.

Parameters:

numLabels - Number of labels to adjust the output

seed - Seed

shape - shape

filterMode - True if creating for feature extraction

Returns:

MultiLayerNetwork of the specified ZooModel

Throws:

java.lang.UnsupportedOperationException - Init(...) was not supported (only CustomNet)

getDefaultGraph

public org.deeplearning4j.nn.graph.ComputationGraph getDefaultGraph()

Convenience method to returns a default pretrained graph for this zoo model

Returns:

getInputShape

public abstract int[] getInputShape()

Get the input shape of this zoomodel

Returns:

Input shape of this zoomodel

getPrettyName

public java.lang.String getPrettyName()

getVariation

public abstract java.lang.Enum getVariation()

Get the current variation of the zoo model (e.g., Resnet50 or Resnet101)

Returns:

Variation

requiresPreProcessing

public boolean requiresPreProcessing()

Does the model require input images to be preprocessed?.

Returns:

true if the model input should be rescaled

getImagePreprocessingScaler

public abstract org.nd4j.linalg.dataset.api.preprocessor.ImagePreProcessingScaler getImagePreprocessingScaler()

Get the preprocessor to process this model's data with

Returns:

DataSetPreprocessor

getChannelsLast

@OptionMetadata(displayName="Image channels last",
                description="Set to true to supply image channels last. The default value will usually be correct, so as an end user you shouldn\'t need to change this setting. If you do be aware that it may break the model.",
                commandLineParamName="channelsLast",
                commandLineParamSynopsis="-channelsLast <boolean>")
public boolean getChannelsLast()

setChannelsLast

public void setChannelsLast(boolean channelsLast)

isRequiresPooling

@ProgrammaticProperty
public boolean isRequiresPooling()

setRequiresPooling

public void setRequiresPooling(boolean requiresPooling)

initZooModel

public org.deeplearning4j.nn.graph.ComputationGraph initZooModel(org.deeplearning4j.zoo.ZooModel zooModel,
                                                                 org.deeplearning4j.nn.graph.ComputationGraph defaultNet,
                                                                 long seed,
                                                                 int numLabels,
                                                                 boolean filterMode)

Initialize the zoo model with the supplied params.

Parameters:

zooModel - Zoo model family to use

defaultNet - Default ComputationGraph to use if loading weights fails

seed - Random seed to initialize with

numLabels - Number of output labels

filterMode - True if using this zoo model for a filter - output layers don't need to be setup

Returns:

ComputationGraph - if all succeeds then will be initialized with pretrained weights

isPretrained

public boolean isPretrained()

Returns:

True if current model is pretrained

getPretrainedType

@OptionMetadata(description="Pretrained Type (IMAGENET, VGGFACE, MNIST)",
                displayName="Pretrained Type",
                commandLineParamName="pretrained",
                commandLineParamSynopsis="-pretrained <string>")
public PretrainedType getPretrainedType()

setPretrainedType

public void setPretrainedType(PretrainedType pretrainedType)

getOutputlayer

@ProgrammaticProperty
public java.lang.String getOutputlayer()

setOutputLayer

public void setOutputLayer(java.lang.String m_outputLayer)

getFeatureExtractionLayer

@ProgrammaticProperty
public java.lang.String getFeatureExtractionLayer()

setFeatureExtractionLayer

public void setFeatureExtractionLayer(java.lang.String m_featureExtractionLayer)

getExtraLayersToRemove

@ProgrammaticProperty
public java.lang.String[] getExtraLayersToRemove()

setExtraLayersToRemove

public void setExtraLayersToRemove(java.lang.String[] m_extraLayersToRemove)

getNumFExtractOutputs

@ProgrammaticProperty
public int getNumFExtractOutputs()

setNumFExtractOutputs

public void setNumFExtractOutputs(int m_numFExtractOutputs)

listOptions

public java.util.Enumeration<weka.core.Option> listOptions()

Returns an enumeration describing the available options.

Specified by:

listOptions in interface weka.core.OptionHandler

Returns:

an enumeration of all the available options.

getOptions

public java.lang.String[] getOptions()

Gets the current settings of the Classifier.

Specified by:

getOptions in interface weka.core.OptionHandler

Returns:

an array of strings suitable for passing to setOptions

setOptions

public void setOptions(java.lang.String[] options)
                throws java.lang.Exception

Parses a given list of options.

Specified by:

setOptions in interface weka.core.OptionHandler

Parameters:

options - the list of options as an array of strings

Throws:

java.lang.Exception - if an option is not supported