org.deeplearning4j.nn.layers.convolution

Class SeparableConvolution2DLayer

java.lang.Object

org.deeplearning4j.nn.layers.AbstractLayer<LayerConfT>

org.deeplearning4j.nn.layers.BaseLayer<ConvolutionLayer>

org.deeplearning4j.nn.layers.convolution.ConvolutionLayer

org.deeplearning4j.nn.layers.convolution.SeparableConvolution2DLayer

All Implemented Interfaces:

Serializable, Cloneable, Layer, Model, Trainable

public class SeparableConvolution2DLayer
extends ConvolutionLayer

2D Separable convolution layer implementation Separable convolutions split a regular convolution operation into two simpler operations, which are usually computationally more efficient. The first step in a separable convolution is a channels-wise convolution, which operates on each of the input maps separately. A channels multiplier is used to specify the number of outputs per input map in this step. This convolution is carried out with the specified kernel sizes, stride and padding values. The second step is a point-wise operation, in which the intermediary outputs of the channels-wise convolution are mapped to the desired number of feature maps, by using a 1x1 convolution. The result of chaining these two operations will result in a tensor of the same shape as that for a standard conv2d operation.

Author:

Max Pumperla

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from interface org.deeplearning4j.nn.api.Layer

Layer.TrainingMode, Layer.Type

Field Summary

Fields inherited from class org.deeplearning4j.nn.layers.convolution.ConvolutionLayer

convolutionMode, dummyBias, dummyBiasGrad, helper, helperCountFail, i2d, log

Fields inherited from class org.deeplearning4j.nn.layers.BaseLayer

gradient, gradientsFlattened, gradientViews, optimizer, params, paramsFlattened, score, solver, weightNoiseParams

Fields inherited from class org.deeplearning4j.nn.layers.AbstractLayer

cacheMode, conf, dataType, dropoutApplied, epochCount, index, input, inputModificationAllowed, iterationCount, maskArray, maskState, preOutput, trainingListeners

Constructor Summary

Constructors

Constructor and Description
SeparableConvolution2DLayer(NeuralNetConfiguration conf, org.nd4j.linalg.api.buffer.DataType dataType)

Method Summary

Modifier and Type	Method and Description
INDArray	activate(boolean training, LayerWorkspaceMgr workspaceMgr) Perform forward pass and return the activations array with the last set input
Pair<Gradient,INDArray>	backpropGradient(INDArray epsilon, LayerWorkspaceMgr workspaceMgr) Calculate the gradient relative to the error in the next layer
protected Pair<INDArray,INDArray>	preOutput(boolean training, boolean forBackprop, LayerWorkspaceMgr workspaceMgr) PreOutput method that also returns the im2col2d array (if being called for backprop), as this can be re-used instead of being calculated again.

Methods inherited from class org.deeplearning4j.nn.layers.convolution.ConvolutionLayer

feedForwardMaskArray, fit, getHelper, hasBias, isPretrainLayer, preOutput4d, setParams, type, validateInputDepth, validateInputRank

Methods inherited from class org.deeplearning4j.nn.layers.BaseLayer

calcRegularizationScore, clear, clearNoiseWeightParams, clone, computeGradientAndScore, fit, getGradientsViewArray, getOptimizer, getParam, getParamWithNoise, gradient, hasLayerNorm, layerConf, numParams, params, paramTable, paramTable, preOutput, preOutputWithPreNorm, score, setBackpropGradientsViewArray, setParam, setParams, setParamsViewArray, setParamTable, setScoreWithZ, toString, update, update

Methods inherited from class org.deeplearning4j.nn.layers.AbstractLayer

activate, addListeners, allowInputModification, applyConstraints, applyDropOutIfNecessary, applyMask, assertInputSet, backpropDropOutIfPresent, batchSize, conf, getConfig, getEpochCount, getIndex, getInput, getInputMiniBatchSize, getListeners, getMaskArray, gradientAndScore, init, input, layerId, numParams, setCacheMode, setConf, setEpochCount, setIndex, setInput, setInputMiniBatchSize, setListeners, setListeners, setMaskArray, updaterDivideByMinibatch

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface org.deeplearning4j.nn.api.Layer

getIterationCount, setIterationCount

Constructor Detail

SeparableConvolution2DLayer

public SeparableConvolution2DLayer(NeuralNetConfiguration conf,
                                   org.nd4j.linalg.api.buffer.DataType dataType)

Method Detail

backpropGradient

public Pair<Gradient,INDArray> backpropGradient(INDArray epsilon,
                                                LayerWorkspaceMgr workspaceMgr)

Description copied from interface: Layer

Calculate the gradient relative to the error in the next layer

Specified by:

backpropGradient in interface Layer

Overrides:

backpropGradient in class ConvolutionLayer

Parameters:

epsilon - w^(L+1)*delta^(L+1). Or, equiv: dC/da, i.e., (dC/dz)*(dz/da) = dC/da, where C is cost function a=sigma(z) is activation.

workspaceMgr - Workspace manager

Returns:

Pair where Gradient is gradient for this layer, INDArray is epsilon (activation gradient) needed by next layer, but before element-wise multiply by sigmaPrime(z). So for standard feed-forward layer, if this layer is L, then return.getSecond() == dL/dIn = (w^(L)*(delta^(L))^T)^T. Note that the returned array should be placed in the ArrayType.ACTIVATION_GRAD workspace via the workspace manager

preOutput

protected Pair<INDArray,INDArray> preOutput(boolean training,
                                            boolean forBackprop,
                                            LayerWorkspaceMgr workspaceMgr)

Description copied from class: ConvolutionLayer

PreOutput method that also returns the im2col2d array (if being called for backprop), as this can be re-used instead of being calculated again.

Overrides:

preOutput in class ConvolutionLayer

Parameters:

training - Train or test time (impacts dropout)

forBackprop - If true: return the im2col2d array for re-use during backprop. False: return null for second pair entry. Note that it may still be null in the case of CuDNN and the like.

Returns:

Pair of arrays: preOutput (activations) and optionally the im2col2d array

activate

public INDArray activate(boolean training,
                         LayerWorkspaceMgr workspaceMgr)

Description copied from interface: Layer

Perform forward pass and return the activations array with the last set input

Specified by:

activate in interface Layer

Overrides:

activate in class ConvolutionLayer

Parameters:

training - training or test mode

workspaceMgr - Workspace manager

Returns:

the activation (layer output) of the last specified input. Note that the returned array should be placed in the ArrayType.ACTIVATIONS workspace via the workspace manager