Tensor Class Reference

#include <ImFusion/ML/Tensor.h>

Implementation of a Tensor class; used in TensorDataElement to directly present data to a neural network. More...

Detailed Description

Implementation of a Tensor class; used in TensorDataElement to directly present data to a neural network.

Currently suppports three datatypes: Float, Int (int64_t) and uint8_t.

Public Member Functions
	Tensor ()
	Create empty Tensor.
	Tensor (Tensor const &other)=default
	Copy constructor.
Tensor &	operator= (const Tensor &other)=default
	Copy assignement.
	Tensor (Tensor &&other) noexcept
	Move constructor.
Tensor &	operator= (Tensor &&other) noexcept
	Move assignement.
	Tensor (std::vector< size_t > shape, float const *data)
	Create a tensor with shape 'shape' and type Float, initialized from data.
	Tensor (std::vector< size_t > shape, int64_t const *data)
	Create a tensor with shape 'shape' and type Int, initialized from data.
	Tensor (std::vector< size_t > shape, uint8_t const *data)
	Create a tensor with shape 'shape' and type uint8_t, initialized from data.
	Tensor (std::vector< size_t > shape, PixelType type, uint8_t const *raw_data=nullptr)
	Create a tensor with shape 'shape' and type 'type', initialized from data.
size_t	order () const
	Rank of the tensor.
std::vector< size_t > const &	shape () const
	Returns the shape of the tensor.
size_t	numElements () const
	Returns the number of elements in this tensor's storage buffer.
size_t	byteSize () const
	Returns the number of bytes in this tensor's storage buffer.
bool	serialize (std::ostream &out) const
	Writes the tensor content into the given stream.
bool	deserialize (std::istream &in)
	Clears the tensor and reads new content from the given stream.
bool	operator== (Tensor const &other) const
std::vector< uint64_t > const &	byteData () const
	Return the raw underlying data as bytes.
PixelType	dataType () const
std::optional< std::vector< float > >	toFloatVector () const
	Create a copy of underlying data into a float vector, of size numElements().
std::optional< std::vector< int64_t > >	toIntVector () const
	Create a copy of underlying data into a int64_t vector, of size numElements().
std::optional< std::vector< uint8_t > >	toUByteVector () const
	Create a copy of underlying data into a uint8_t vector, of size numElements().
uint8_t const *	bytePtr () const
	Access to raw byte data.
uint8_t *	bytePtr ()

Protected Member Functions
void	swapWith (Tensor &other)

Constructor & Destructor Documentation

Tensor() [1/4]

Tensor ( std::vector< size_t > shape, float const * data )

explicit

Create a tensor with shape 'shape' and type Float, initialized from data.

Makes a copy of the data, does not take ownership.

Tensor() [2/4]

Tensor ( std::vector< size_t > shape, int64_t const * data )

explicit

Create a tensor with shape 'shape' and type Int, initialized from data.

Makes a copy of the data, does not take ownership.

Tensor() [3/4]

Tensor ( std::vector< size_t > shape, uint8_t const * data )

explicit

Create a tensor with shape 'shape' and type uint8_t, initialized from data.

Makes a copy of the data, does not take ownership.

Tensor() [4/4]

Tensor ( std::vector< size_t > shape, PixelType type, uint8_t const * raw_data = nullptr )

explicit

Create a tensor with shape 'shape' and type 'type', initialized from data.

Makes a copy of the raw_data if supplied, otherwise initialize as 0. Does not take ownership.

Member Function Documentation

toFloatVector()

std::optional< std::vector< float > > toFloatVector

(

)

const

Create a copy of underlying data into a float vector, of size numElements().

Element ordering follows Torch ordering (lowest dimension is last shape dimension). Returns null optional if data is not of float type.

toIntVector()

std::optional< std::vector< int64_t > > toIntVector

(

)

const

Create a copy of underlying data into a int64_t vector, of size numElements().

Element ordering follows Torch ordering (lowest dimension is last shape dimension). Returns null optional if data is not of int64_t type.

toUByteVector()

std::optional< std::vector< uint8_t > > toUByteVector

(

)

const

Create a copy of underlying data into a uint8_t vector, of size numElements().

Element ordering follows Torch ordering (lowest dimension is last shape dimension). Returns null optional if data is not of uint8_t type.

---

The documentation for this class was generated from the following file:

• ImFusion/ML/Tensor.h