Struct TensorShape

TensorShape are immutable representation of a Tensor dimensions and rank. Depending on which constructor is used, the TensorShape will either be rank 8 and channels last (ie NHWC) or actual rank with unnamed tensor dimensions when using the constructor that takes int[]. With legacy use (explicit named constructors) of TensorShape an axis can be of size 1. For example, a tensor without spatial information will be N,1,1,C. With the use of TensorShape via the int[] constructor, then axes can have values of 0.

Inherited Members

Object.Equals(Object, Object)

Object.ReferenceEquals(Object, Object)

Object.GetType()

Namespace: Unity.Barracuda

Syntax

[Serializable]
public struct TensorShape

Constructors

TensorShape(Int32)

Create a TensorShape of shape [1,1,N,1,1,1,1,1].

Declaration

public TensorShape(int n)

Parameters

Type	Name	Description
Int32	n	batch

TensorShape(Int32, Int32)

Create a TensorShape of shape [1,1,N,1,1,1,1,C].

Declaration

public TensorShape(int n, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	c	channels

TensorShape(Int32, Int32, Int32)

Create a TensorShape of shape [1,1,N,1,1,1,W,C].

Declaration

public TensorShape(int n, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	w	width
Int32	c	channels

TensorShape(Int32, Int32, Int32, Int32)

Create a TensorShape of shape [1,1,N,1,1,H,W,C].

Declaration

public TensorShape(int n, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	h	height
Int32	w	width
Int32	c	channels

TensorShape(Int32, Int32, Int32, Int32, Int32)

Create a TensorShape of shape [1,1,N,1,D,H,W,C].

Declaration

public TensorShape(int n, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

TensorShape(Int32, Int32, Int32, Int32, Int32, Int32, Int32, Int32)

Create a TensorShape of shape [S,R,N,T,D,H,W,C]. Currently seqLen must be 1.

Declaration

public TensorShape(int s, int r, int n, int t, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	s	sequence
Int32	r	direction
Int32	n	batch
Int32	t	time
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

TensorShape(Int32[], Boolean)

Create a TensorShape of arbitrary shape.

Declaration

public TensorShape(int[] shape, bool unnamedDimensions = false)

Parameters

Type	Name	Description
Int32[]	shape	shape as int array
Boolean	unnamedDimensions	create the shape with no specific, named layout

Fields

C

Channels dimension index number

Declaration

public const int C = 7

Field Value

Type	Description
Int32

D

Depth dimension index number

Declaration

public const int D = 4

Field Value

Type	Description
Int32

DataBatch

Batch dimension index number

Declaration

public const int DataBatch = 2

Field Value

Type	Description
Int32

DataChannel

Data channel dimension index number

Declaration

public const int DataChannel = 7

Field Value

Type	Description
Int32

DataFeature0

Data feature 0 dimension index number

Declaration

public const int DataFeature0 = 6

Field Value

Type	Description
Int32

DataFeature1

Data feature 1 dimension index number

Declaration

public const int DataFeature1 = 5

Field Value

Type	Description
Int32

DataFeature2

Data feature 2 dimension index number

Declaration

public const int DataFeature2 = 4

Field Value

Type	Description
Int32

DataFeature3

Data feature 3 dimension index number

Declaration

public const int DataFeature3 = 3

Field Value

Type	Description
Int32

DataFeatures

Data features

Declaration

public static readonly int[] DataFeatures

Field Value

Type	Description
Int32[]

H

Height dimension index number

Declaration

public const int H = 5

Field Value

Type	Description
Int32

KernelInChannel

Kernel input channel dimension

Declaration

public const int KernelInChannel = 6

Field Value

Type	Description
Int32

KernelOutChannel

Kernel output channel dimension

Declaration

public const int KernelOutChannel = 7

Field Value

Type	Description
Int32

KernelSpatial0

Kernel spatial dimension 0

Declaration

public const int KernelSpatial0 = 5

Field Value

Type	Description
Int32

KernelSpatial1

Kernel spatial dimension 1

Declaration

public const int KernelSpatial1 = 2

Field Value

Type	Description
Int32

KernelSpatial2

Kernel spatial dimension 2

Declaration

public const int KernelSpatial2 = 1

Field Value

Type	Description
Int32

KernelSpatial3

Kernel spatial dimension 3

Declaration

public const int KernelSpatial3 = 0

Field Value

Type	Description
Int32

KernelSpatials

Kernel spatial dimensions

Declaration

public static readonly int[] KernelSpatials

Field Value

Type	Description
Int32[]

MaxRank

Max rank

Declaration

public const int MaxRank = 8

Field Value

Type	Description
Int32

NumberOfDirections

Sequence length dimension index number

Declaration

public const int NumberOfDirections = 1

Field Value

Type	Description
Int32

SequenceLength

Sequence length dimension index number

Declaration

public const int SequenceLength = 0

Field Value

Type	Description
Int32

W

Width dimension index number

Declaration

public const int W = 6

Field Value

Type	Description
Int32

Properties

batch

Return the number of batch.

Declaration

public readonly int batch { get; }

Property Value

Type	Description
Int32

channels

Return the number of channels.

Declaration

public readonly int channels { get; }

Property Value

Type	Description
Int32

depth

Return the spatial depth (axis is DataFeature2).

Declaration

public readonly int depth { get; }

Property Value

Type	Description
Int32

dimensions

Return the count of non-unit dimension of this shape. For example [N,1,1,C] dimensions is 2.

Declaration

public readonly int dimensions { get; }

Property Value

Type	Description
Int32

flatHeight

Return the number of batch.

Declaration

public readonly int flatHeight { get; }

Property Value

Type	Description
Int32

flatWidth

Return the TDHWC.

Declaration

public readonly int flatWidth { get; }

Property Value

Type	Description
Int32

hasNamedDimensions

Whether this shape makes use of named dimensions or is nameless.

Declaration

public readonly bool hasNamedDimensions { get; }

Property Value

Type	Description
Boolean

height

Return the spatial height (axis is DataFeature1).

Declaration

public readonly int height { get; }

Property Value

Type	Description
Int32

Item[Int32]

Indexer to return a dimension of this tensorShape as [S,R,N,T,D,H,W,C] Prefer this over ToArray() to avoid GC allocation/collection.

Declaration

public int this[int axis] { get; }

Parameters

Type	Name	Description
Int32	axis	axis

Property Value

Type	Description
Int32

kernelCount

Kernel dimension ordering is [D,H,W,C,K] for efficiency purpose. Return kernel count (aka the number of output channels of the associated operator).

Declaration

public readonly int kernelCount { get; }

Property Value

Type	Description
Int32

kernelDepth

Kernel dimension ordering is [D,H,W,C,K] for efficiency purpose. Return kernel depth (aka the number of input channels of the associated operator).

Declaration

public readonly int kernelDepth { get; }

Property Value

Type	Description
Int32

kernelHeight

Kernel dimension ordering is [D,H,W,C,K] for efficiency purpose. Return kernel height.

Declaration

public readonly int kernelHeight { get; }

Property Value

Type	Description
Int32

kernelSpatialDepth

Kernel dimension ordering is [D,H,W,C,K] for efficiency purpose. Return kernel intermediate dimension 0.

Declaration

public readonly int kernelSpatialDepth { get; }

Property Value

Type	Description
Int32

kernelWidth

Kernel dimension ordering is [D,H,W,C,K] for efficiency purpose. Return kernel width.

Declaration

public readonly int kernelWidth { get; }

Property Value

Type	Description
Int32

length

Return the total number of elements represented by this shape.

Declaration

public readonly int length { get; }

Property Value

Type	Description
Int32

numberOfDirections

Return the number of direction.

Declaration

public readonly int numberOfDirections { get; }

Property Value

Type	Description
Int32

rank

Always 8 if legacy, named constructors are used otherwise the actual rank. Look also at the dimensions property.

Declaration

public readonly int rank { get; }

Property Value

Type	Description
Int32

sequenceLength

Return the number of sequence.

Declaration

public readonly int sequenceLength { get; }

Property Value

Type	Description
Int32

width

Return the spatial width (axis is DataFeature0).

Declaration

public readonly int width { get; }

Property Value

Type	Description
Int32

Methods

AsNamed()

Declaration

public TensorShape AsNamed()

Returns

Type	Description
TensorShape

AsUnnamed()

Declaration

public TensorShape AsUnnamed()

Returns

Type	Description
TensorShape

Axis(Int32)

Allow to use negative axis to access tensorShape backward. axis should be from -rank to rank (exclusive).

Declaration

public int Axis(int axis)

Parameters

Type	Name	Description
Int32	axis	axis

Returns

Type	Description
Int32	adjusted axis

Equals(Object)

Compares this object to other object

Declaration

public override bool Equals(object obj)

Parameters

Type	Name	Description
Object	obj	other object

Returns

Type	Description
Boolean	true if contents of the objects a and b are equal

Overrides

ValueType.Equals(Object)

Flatten()

Return a TensorShape of dimensions [S,R,N,1,1,1,1,TDHWC]

Declaration

public TensorShape Flatten()

Returns

Type	Description
TensorShape	new TensorShape

GetHashCode()

Object hash code

Declaration

public override int GetHashCode()

Returns

Type	Description
Int32	object hash code

Overrides

ValueType.GetHashCode()

GetPositionsFromIndex(Int32, ref Int32, ref Int32, ref Int32, ref Int32)

Given an offset in memory return the dimensions indices of the element as [,,N,,,H,W,C].

Declaration

public void GetPositionsFromIndex(int index, ref int n, ref int h, ref int w, ref int c)

Parameters

Type	Name	Description
Int32	index	one dimensional index (offset) in the memory
Int32	n	batch
Int32	h	height
Int32	w	width
Int32	c	channels

GetPositionsFromIndex(Int32, ref Int32, ref Int32, ref Int32, ref Int32, ref Int32, ref Int32, ref Int32, ref Int32)

Given an offset in memory return the dimensions indices of the element as [S,R,N,T,D,H,W,C].

Declaration

public void GetPositionsFromIndex(int index, ref int s, ref int r, ref int n, ref int t, ref int d, ref int h, ref int w, ref int c)

Parameters

Type	Name	Description
Int32	index	one dimensional index (offset) in the memory
Int32	s	sequence
Int32	r	direction
Int32	n	batch
Int32	t	time
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Index(Int32, Int32)

Given an element dimensions indices [0,0,N,0,0,0,0,C] return this element offset in memory.

Declaration

public int Index(int n, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

Index(Int32, Int32, Int32, Int32)

Given an element dimensions indices [0,0,N,0,0,H,W,C] return this element offset in memory.

Declaration

public int Index(int n, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

Index(Int32, Int32, Int32, Int32, Int32)

Given an element dimensions indices [0,0,N,0,D,H,W,C] return this element offset in memory.

Declaration

public int Index(int n, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

Index(Int32, Int32, Int32, Int32, Int32, Int32, Int32, Int32)

Given an element dimensions indices [S,R,N,T,D,H,W,C] return this element offset in memory.

Declaration

public int Index(int s, int r, int n, int t, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	s	sequence
Int32	r	direction
Int32	n	batch
Int32	t	time
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

IndexWithBroadcast(Int32, Int32, Int32, Int32)

Given an element dimensions indices [0,0,N,0,0,H,W,C] with broadcast support, return this element offset in memory.

Declaration

public int IndexWithBroadcast(int n, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32

IndexWithBroadcast(Int32, Int32, Int32, Int32, Int32, Int32, Int32, Int32)

Given an element dimensions indices [S,R,N,T,D,H,W,C] with broadcast support, return this element offset in memory.

Declaration

public int IndexWithBroadcast(int s, int r, int n, int t, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	s	sequence
Int32	r	direction
Int32	n	batch
Int32	t	time
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

IndexWithClamp(Int32, Int32, Int32, Int32)

Given an element dimensions indices [1,N,1,1,1,H,W,C] return this element offset in memory, clamping indices to tensor dimensions.

Declaration

public int IndexWithClamp(int n, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

IndexWithClamp(Int32, Int32, Int32, Int32, Int32)

Given an element dimensions indices [1,N,1,1,D,H,W,C] return this element offset in memory, clamping indices to tensor dimensions.

Declaration

public int IndexWithClamp(int n, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	n	batch
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

IndexWithClamp(Int32, Int32, Int32, Int32, Int32, Int32, Int32, Int32)

Given an element dimensions indices [S,R,N,T,D,H,W,C] return this element offset in memory, clamping indices to tensor dimensions.

Declaration

public int IndexWithClamp(int s, int r, int n, int t, int d, int h, int w, int c)

Parameters

Type	Name	Description
Int32	s	sequence
Int32	r	direction
Int32	n	batch
Int32	t	time
Int32	d	depth
Int32	h	height
Int32	w	width
Int32	c	channels

Returns

Type	Description
Int32	one dimensional index (offset in the flat memory region)

Squeeze()

Remove single-dimensional entries from the shape. [s=1,r=1,b=4,t=1,d=1h=1,w=1,c=128] => [s=1,r=1,b=1,t=1,d=1,h=1,w=4,c=128]

Declaration

public TensorShape Squeeze()

Returns

Type	Description
TensorShape	new TensorShape

ToArray()

Return an array representation of this tensorShape as [S,R,N,T,D,H,W,C] Prefer tensorShape[x] to avoid GC allocation/collection.

Declaration

public int[] ToArray()

Returns

Type	Description
Int32[]	shape as int array

ToString()

Object summary

Declaration

public override string ToString()

Returns

Type	Description
String	object summary as a string

Overrides

ValueType.ToString()

Operators

Equality(TensorShape, TensorShape)

Compares two TensorShape objects

Declaration

public static bool operator ==(TensorShape a, TensorShape b)

Parameters

Type	Name	Description
TensorShape	a	left object
TensorShape	b	right object

Returns

Type	Description
Boolean	true if contents of the objects a and b are equal

Inequality(TensorShape, TensorShape)

Compares two TensorShape objects

Declaration

public static bool operator !=(TensorShape a, TensorShape b)

Parameters

Type	Name	Description
TensorShape	a	left object
TensorShape	b	right object

Returns

Type	Description
Boolean	true if contents of the objects a and b are not equal

Extension Methods

DeprecatedTensorExtensions.AdjustPadToPool(TensorShape, (Int32, Int32), Int32[], Int32[])

TensorExtensions.Scale(TensorShape, TensorShape)

TensorExtensions.Scale(TensorShape, Int32[])

TensorExtensions.Reduce(TensorShape, Int32)

TensorExtensions.Reshape(TensorShape, Int32[])

TensorExtensions.ApplyBorder(TensorShape, Int32[])

TensorExtensions.Get8DPermutationsForNHWCPermutationsAndShape(TensorShape, Int32[])

TensorExtensions.Permute(TensorShape, Int32[])