org.deeplearning4j.util

Class Convolution3DUtils

java.lang.Object

org.deeplearning4j.util.Convolution3DUtils

public class Convolution3DUtils
extends Object

Method Summary

Modifier and Type	Method and Description
static int[]	get3DOutputSize(INDArray inputData, int[] kernel, int[] strides, int[] padding, ConvolutionMode convolutionMode, int[] dilation, boolean isNCDHW) Get the output size (depth/height/width) for the given input data and CNN3D configuration
static int[]	get3DSameModeTopLeftPadding(int[] outSize, int[] inSize, int[] kernel, int[] strides, int[] dilation) Get top and left padding for same mode only for 3d convolutions
static Convolution3D.DataFormat	getFormatForLayer(Layer inputLayer) Returns the Convolution3D.DataFormat for the associated layer.
static boolean	layerHasConvolution3DLayout(Layer layer) Returns true if any of the layers are 3d convolution, pooling, or upsampling layers including: Convolution3D, Deconvolution3D, Subsampling3DLayer, Upsampling3D
static void	validateCnn3DKernelStridePadding(int[] kernelSize, int[] stride, int[] padding) Perform validation on the CNN3D layer kernel/stride/padding.

Methods inherited from class java.lang.Object

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

Method Detail

get3DOutputSize

public static int[] get3DOutputSize(INDArray inputData,
                                    int[] kernel,
                                    int[] strides,
                                    int[] padding,
                                    ConvolutionMode convolutionMode,
                                    int[] dilation,
                                    boolean isNCDHW)

Get the output size (depth/height/width) for the given input data and CNN3D configuration

Parameters:

inputData - Input data

kernel - Kernel size (depth/height/width)

strides - Strides (depth/height/width)

padding - Padding (depth/height/width)

convolutionMode - Convolution mode (Same, Strict, Truncate)

dilation - Kernel dilation (depth/height/width)

Returns:

Output size: int[3] with output depth/height/width

get3DSameModeTopLeftPadding

public static int[] get3DSameModeTopLeftPadding(int[] outSize,
                                                int[] inSize,
                                                int[] kernel,
                                                int[] strides,
                                                int[] dilation)

Get top and left padding for same mode only for 3d convolutions

Parameters:

outSize -

inSize -

kernel -

strides -

Returns:

validateCnn3DKernelStridePadding

public static void validateCnn3DKernelStridePadding(int[] kernelSize,
                                                    int[] stride,
                                                    int[] padding)

Perform validation on the CNN3D layer kernel/stride/padding. Expect 3d int[], with values > 0 for kernel size and stride, and values >= 0 for padding.

Parameters:

kernelSize - Kernel size array to check

stride - Stride array to check

padding - Padding array to check

getFormatForLayer

public static Convolution3D.DataFormat getFormatForLayer(Layer inputLayer)

Returns the Convolution3D.DataFormat for the associated layer. Note that layerHasConvolution3DLayout(Layer) should return true on the given layer input or an IllegalArgumentException will be thrown

Parameters:

inputLayer - the input layer to get the format for

Returns:

the associated Convolution3D.DataFormat with the layer

layerHasConvolution3DLayout

public static boolean layerHasConvolution3DLayout(Layer layer)

Returns true if any of the layers are 3d convolution, pooling, or upsampling layers including: Convolution3D, Deconvolution3D, Subsampling3DLayer, Upsampling3D

Parameters:

layer - the input layer to validate

Returns:

true if the layer is any of the types specified.