DepthToSpace

public final class DepthToSpace

DepthToSpace for tensors of type T.

Rearranges data from depth into blocks of spatial data. This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of the input tensor where values from the `depth` dimension are moved in spatial blocks to the `height` and `width` dimensions. The attr `block_size` indicates the input block size and how the data is moved.

* Chunks of data of size `block_size * block_size` from depth are rearranged into non-overlapping blocks of size `block_size x block_size` * The width the output tensor is `input_depth * block_size`, whereas the height is `input_height * block_size`. * The Y, X coordinates within each block of the output image are determined by the high order component of the input channel index. * The depth of the input tensor must be divisible by `block_size * block_size`.

The `data_format` attr specifies the layout of the input and output tensors with the following options: "NHWC": `[ batch, height, width, channels ]` "NCHW": `[ batch, channels, height, width ]` "NCHW_VECT_C": `qint8 [ batch, channels / 4, height, width, 4 ]`

It is useful to consider the operation as transforming a 6-D Tensor. e.g. for data_format = NHWC, Each element in the input tensor can be specified via 6 coordinates, ordered by decreasing memory layout significance as: n,iY,iX,bY,bX,oC (where n=batch index, iX, iY means X or Y coordinates within the input image, bX, bY means coordinates within the output block, oC means output channels). The output would be the input transposed to the following layout: n,iY,bY,iX,bX,oC

This operation is useful for resizing the activations between convolutions (but keeping all data), e.g. instead of pooling. It is also useful for training purely convolutional models.

For example, given an input of shape `[1, 1, 1, 4]`, data_format = "NHWC" and block_size = 2:

x = [[[[1, 2, 3, 4]]]]
 
 

This operation will output a tensor of shape `[1, 2, 2, 1]`:

[[[[1], [2]],
      [[3], [4]]]]
 

Here, the input has a batch of 1 and each batch element has shape `[1, 1, 4]`, the corresponding output will have 2x2 elements and will have a depth of 1 channel (1 = `4 / (block_size * block_size)`). The output element shape is `[2, 2, 1]`.

For an input tensor with larger depth, here of shape `[1, 1, 1, 12]`, e.g.

x = [[[[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]]]]
 

This operation, for block size of 2, will return the following tensor of shape `[1, 2, 2, 3]`

[[[[1, 2, 3], [4, 5, 6]],
      [[7, 8, 9], [10, 11, 12]]]]
 
 

Similarly, for the following input of shape `[1 2 2 4]`, and a block size of 2:

x =  [[[[1, 2, 3, 4],
        [5, 6, 7, 8]],
       [[9, 10, 11, 12],
        [13, 14, 15, 16]]]]
 

the operator will return the following tensor of shape `[1 4 4 1]`:

x = [[[ [1],   [2],  [5],  [6]],
       [ [3],   [4],  [7],  [8]],
       [ [9],  [10], [13],  [14]],
       [ [11], [12], [15],  [16]]]]
 
 

Nested Classes

class

DepthToSpace.Options

Optional attributes for DepthToSpace

Constants

String

OP_NAME

The name of this op, as known by TensorFlow core engine

Public Methods

Output<T>	asOutput() Returns the symbolic handle of the tensor.
static <T extends TType> DepthToSpace<T>	create(Scope scope, Operand<T> input, Long blockSize, Options... options) Factory method to create a class wrapping a new DepthToSpace operation.
static DepthToSpace.Options	dataFormat(String dataFormat)
Output<T>	output()

Inherited Methods

From class org.tensorflow.op.RawOp

final boolean	equals(Object obj)
final int	hashCode()
Operation	op() Return this unit of computation as a single Operation.
final String	toString()

From class java.lang.Object

boolean	equals(Object arg0)
final Class<?>	getClass()
int	hashCode()
final void	notify()
final void	notifyAll()
String	toString()
final void	wait(long arg0, int arg1)
final void	wait(long arg0)
final void	wait()

From interface org.tensorflow.op.Op

abstract ExecutionEnvironment	env() Return the execution environment this op was created in.
abstract Operation	op() Return this unit of computation as a single Operation.

From interface org.tensorflow.Operand

abstract Output<T>	asOutput() Returns the symbolic handle of the tensor.
abstract T	asTensor() Returns the tensor at this operand.
abstract Shape	shape() Returns the (possibly partially known) shape of the tensor referred to by the Output of this operand.
abstract Class<T>	type() Returns the tensor type of this operand

From interface org.tensorflow.ndarray.Shaped

abstract int	rank()
abstract Shape	shape()
abstract long	size() Computes and returns the total size of this container, in number of values.

Constants

public static final String OP_NAME

The name of this op, as known by TensorFlow core engine

Constant Value: "DepthToSpace"

Public Methods

public Output<T> asOutput ()

Returns the symbolic handle of the tensor.

Inputs to TensorFlow operations are outputs of another TensorFlow operation. This method is used to obtain a symbolic handle that represents the computation of the input.

public static DepthToSpace<T> create (Scope scope, Operand<T> input, Long blockSize, Options... options)

Factory method to create a class wrapping a new DepthToSpace operation.

Parameters

scope	current scope
blockSize	The size of the spatial block, same as in Space2Depth.
options	carries optional attributes values

Returns

• a new instance of DepthToSpace

public static DepthToSpace.Options dataFormat (String dataFormat)

public Output<T> output ()