deepboof.impl.backward.standard

Class DSpatialConvolve2D_F64

java.lang.Object

deepboof.impl.forward.standard.BaseFunction<T>

deepboof.impl.forward.standard.BaseSpatialWindow<T,P>

deepboof.impl.forward.standard.SpatialWindowImage<T,P>

deepboof.impl.backward.standard.DSpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

deepboof.impl.backward.standard.DSpatialConvolve2D_F64

All Implemented Interfaces:

DSpatialConvolve2D<deepboof.tensors.Tensor_F64>, deepboof.DFunction<deepboof.tensors.Tensor_F64>, deepboof.forward.SpatialConvolve2D<deepboof.tensors.Tensor_F64>, deepboof.Function<deepboof.tensors.Tensor_F64>

public class DSpatialConvolve2D_F64
extends DSpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>
implements DSpatialConvolve2D<deepboof.tensors.Tensor_F64>

Implementation of DSpatialConvolve2D for Tensor_F64.

Local caches of the spatial tensor are used to reduce cache misses. The cache will contain a region across all of the tensor's channels that encompasses the region that a single convolution would interact with. In the backwards pass the local cache is written back into the derivative padded input tensor.

Field Summary

Fields

Modifier and Type	Field and Description
protected deepboof.tensors.Tensor_F64	bias
protected double[]	cachedDPadding
protected double[]	cachedPadded
protected deepboof.tensors.Tensor_F64	dBias
protected deepboof.tensors.Tensor_F64	dout
protected deepboof.tensors.Tensor_F64	dWeights
protected int	F
protected deepboof.tensors.Tensor_F64	weights

Fields inherited from class deepboof.impl.backward.standard.DSpatialWindowImage

dpadding, learningMode

Fields inherited from class deepboof.impl.forward.standard.SpatialWindowImage

output

Fields inherited from class deepboof.impl.forward.standard.BaseSpatialWindow

C, config, H, HH, Ho, Hp, N, padding, W, Wo, Wp, WW

Fields inherited from class deepboof.impl.forward.standard.BaseFunction

miniBatchSize, parameters, shapeInput, shapeOutput, shapeParameters

Constructor Summary

Constructors

Constructor and Description
DSpatialConvolve2D_F64(deepboof.forward.ConfigConvolve2D config, DSpatialPadding2D_F64 padding)

Method Summary

Modifier and Type	Method and Description
protected void	_backwards(deepboof.tensors.Tensor_F64 input, deepboof.tensors.Tensor_F64 dout, deepboof.tensors.Tensor_F64 gradientInput, java.util.List<deepboof.tensors.Tensor_F64> gradientParameters)
void	_forward(deepboof.tensors.Tensor_F64 input, deepboof.tensors.Tensor_F64 output)
void	_initialize()
void	_setParameters(java.util.List<deepboof.tensors.Tensor_F64> parameters)
protected void	backwardsAt_border(DSpatialPadding2D_F64 padded, int batch, int padY, int padX, int outY, int outX) Applies the operations at the specified window and stores the results at the specified output coordinate.
protected void	backwardsAt_inner(deepboof.tensors.Tensor_F64 input, int batch, int inY, int inX, int outY, int outX) Applies the operations at the specified window and stores the results at the specified output coordinate.
protected void	forwardAt_border(DSpatialPadding2D_F64 padded, int batch, int padY, int padX, int outY, int outX)
protected void	forwardAt_inner(deepboof.tensors.Tensor_F64 input, int batch, int inY, int inX, int outY, int outX)
deepboof.forward.ConfigConvolve2D	getConfiguration()

Methods inherited from class deepboof.impl.backward.standard.DSpatialWindowImage

backwards, backwardsImage, evaluating, getTensorType, isLearning, learning

Methods inherited from class deepboof.impl.forward.standard.SpatialWindowImage

forwardImage

Methods inherited from class deepboof.impl.forward.standard.BaseSpatialWindow

getPadding, innerLowerExtent, innerUpperExtent, isEntirelyBorder

Methods inherited from class deepboof.impl.forward.standard.BaseFunction

forward, getOutputShape, getParameters, getParameterShapes, initialize, setParameters

Methods inherited from class java.lang.Object

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

Methods inherited from interface deepboof.forward.SpatialConvolve2D

forward, getPadding, setParameters

Methods inherited from interface deepboof.DFunction

backwards, evaluating, isLearning, learning

Methods inherited from interface deepboof.Function

getOutputShape, getParameters, getParameterShapes, getTensorType, initialize

Field Detail

F

protected int F

weights

protected deepboof.tensors.Tensor_F64 weights

bias

protected deepboof.tensors.Tensor_F64 bias

dWeights

protected deepboof.tensors.Tensor_F64 dWeights

dBias

protected deepboof.tensors.Tensor_F64 dBias

dout

protected deepboof.tensors.Tensor_F64 dout

cachedPadded

protected double[] cachedPadded

cachedDPadding

protected double[] cachedDPadding

Constructor Detail

DSpatialConvolve2D_F64

public DSpatialConvolve2D_F64(deepboof.forward.ConfigConvolve2D config,
                              DSpatialPadding2D_F64 padding)

Method Detail

_setParameters

public void _setParameters(java.util.List<deepboof.tensors.Tensor_F64> parameters)

Specified by:

_setParameters in class deepboof.impl.forward.standard.BaseFunction<deepboof.tensors.Tensor_F64>

_initialize

public void _initialize()

Overrides:

_initialize in class deepboof.impl.forward.standard.BaseSpatialWindow<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

_forward

public void _forward(deepboof.tensors.Tensor_F64 input,
                     deepboof.tensors.Tensor_F64 output)

Specified by:

_forward in class deepboof.impl.forward.standard.BaseFunction<deepboof.tensors.Tensor_F64>

forwardAt_inner

protected void forwardAt_inner(deepboof.tensors.Tensor_F64 input,
                               int batch,
                               int inY,
                               int inX,
                               int outY,
                               int outX)

Specified by:

forwardAt_inner in class deepboof.impl.forward.standard.SpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

forwardAt_border

protected void forwardAt_border(DSpatialPadding2D_F64 padded,
                                int batch,
                                int padY,
                                int padX,
                                int outY,
                                int outX)

Specified by:

forwardAt_border in class deepboof.impl.forward.standard.SpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

_backwards

protected void _backwards(deepboof.tensors.Tensor_F64 input,
                          deepboof.tensors.Tensor_F64 dout,
                          deepboof.tensors.Tensor_F64 gradientInput,
                          java.util.List<deepboof.tensors.Tensor_F64> gradientParameters)

Specified by:

_backwards in class DSpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

backwardsAt_inner

protected void backwardsAt_inner(deepboof.tensors.Tensor_F64 input,
                                 int batch,
                                 int inY,
                                 int inX,
                                 int outY,
                                 int outX)

Description copied from class: DSpatialWindowImage

Applies the operations at the specified window and stores the results at the specified output coordinate.

Specified by:

backwardsAt_inner in class DSpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

Parameters:

input - Input spatial tensor

batch - Index of input in mini-batch that is being processed

inY - y-axis lower extent, in input coordinates

inX - x-axis lower extent, in input coordinates

outY - y-axis output coordinates

outX - x-axis output coordinates

backwardsAt_border

protected void backwardsAt_border(DSpatialPadding2D_F64 padded,
                                  int batch,
                                  int padY,
                                  int padX,
                                  int outY,
                                  int outX)

Description copied from class: DSpatialWindowImage

Applies the operations at the specified window and stores the results at the specified output coordinate. For virtual tensor

Specified by:

backwardsAt_border in class DSpatialWindowImage<deepboof.tensors.Tensor_F64,DSpatialPadding2D_F64>

Parameters:

padded - Padded input spatial virtual tensor

batch - Index of input in mini-batch that is being processed

padY - y-axis lower extent, inclusive. Padded coordinates

padX - x-axis lower extent, inclusive. Padded coordinates

outY - y-axis output coordinates

outX - x-axis output coordinates

getConfiguration

public deepboof.forward.ConfigConvolve2D getConfiguration()

Specified by:

getConfiguration in interface deepboof.forward.SpatialConvolve2D<deepboof.tensors.Tensor_F64>