#include <Tensor.h>

Inheritance diagram for open3d::core::Tensor:

Data Structures
struct	ConstIterator
	Const iterator for Tensor. More...

struct	Iterator
	Iterator for Tensor. More...

Public Member Functions
	Tensor ()

	Tensor (const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Constructor for creating a contiguous Tensor. More...

template<typename T >
	Tensor (const std::vector< T > &init_vals, const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Constructor for creating a contiguous Tensor with initial values. More...

template<typename T >
	Tensor (const T *init_vals, const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Constructor from raw host buffer. The memory will be copied. More...

	Tensor (const SizeVector &shape, const SizeVector &strides, void *data_ptr, Dtype dtype, const std::shared_ptr< Blob > &blob)

template<typename T >
	Tensor (std::vector< T > &&vec, const SizeVector &shape={})
	Take ownership of data in std::vector<T> More...

	Tensor (void *data_ptr, Dtype dtype, const SizeVector &shape, const SizeVector &strides={}, const Device &device=Device("CPU:0"))
	Tensor wrapper constructor from raw host buffer. More...

	Tensor (const Tensor &other)=default

	Tensor (Tensor &&other)=default

Tensor &	operator= (const Tensor &other) &

Tensor &	operator= (Tensor &&other) &

Tensor &	operator= (const Tensor &other) &&
	Tensor assignment rvalue = lvalue, e.g. `tensor_a[0] = tensor_b`. More...

Tensor &	operator= (Tensor &&other) &&
	Tensor assignment rvalue = rvalue, e.g. `tensor_a[0] = tensor_b[0]`. More...

template<typename T >
Tensor &	operator= (const T v) &&

Tensor	ReinterpretCast (const core::Dtype &dtype) const

template<typename Object >
Tensor &	AssignObject (const Object &v) &&

template<typename S >
void	Fill (S v)
	Fill the whole Tensor with a scalar value, the scalar will be casted to the Tensor's Dtype. More...

template<typename Object >
void	FillObject (const Object &v)

Tensor	Reverse () const
	Reverse a Tensor's elements by viewing the tensor as a 1D array. More...

Tensor	GetItem (const TensorKey &tk) const

Tensor	GetItem (const std::vector< TensorKey > &tks) const

Tensor	SetItem (const Tensor &value)
	Set all items. Equivalent to `tensor[:] = value` in Python. More...

Tensor	SetItem (const TensorKey &tk, const Tensor &value)

Tensor	SetItem (const std::vector< TensorKey > &tks, const Tensor &value)

Tensor	Append (const Tensor &other, const utility::optional< int64_t > &axis=utility::nullopt) const
	Appends the `other` tensor, along the given axis and returns a copy of the tensor. The `other` tensors must have same data-type, device, and number of dimensions. All dimensions must be the same, except the dimension along the axis the tensors are to be appended. More...

Tensor	Broadcast (const SizeVector &dst_shape) const
	Broadcast Tensor to a new broadcastable shape. More...

Tensor	Expand (const SizeVector &dst_shape) const

Tensor	Reshape (const SizeVector &dst_shape) const

Tensor	Flatten (int64_t start_dim=0, int64_t end_dim=-1) const

Tensor	View (const SizeVector &dst_shape) const

Tensor	Clone () const
	Copy Tensor to the same device. More...

void	CopyFrom (const Tensor &other)
	Copy Tensor values to current tensor from the source tensor. More...

Tensor	To (Dtype dtype, bool copy=false) const

Tensor	To (const Device &device, bool copy=false) const

Tensor	To (const Device &device, Dtype dtype, bool copy=false) const

std::string	ToString (bool with_suffix=true, const std::string &indent="") const

Tensor	operator[] (int64_t i) const
	Extract the i-th Tensor along the first axis, returning a new view. More...

Tensor	IndexExtract (int64_t dim, int64_t idx) const

Tensor	Slice (int64_t dim, int64_t start, int64_t stop, int64_t step=1) const

Tensor	AsRvalue ()

const Tensor	AsRvalue () const
	Convert to constant rvalue. More...

Tensor	IndexGet (const std::vector< Tensor > &index_tensors) const
	Advanced indexing getter. This will always allocate a new Tensor. More...

void	IndexSet (const std::vector< Tensor > &index_tensors, const Tensor &src_tensor)
	Advanced indexing getter. More...

Tensor	Permute (const SizeVector &dims) const
	Permute (dimension shuffle) the Tensor, returns a view. More...

Tensor	AsStrided (const SizeVector &new_shape, const SizeVector &new_strides) const
	Create a Tensor view of specified shape and strides. The underlying buffer and data_ptr offsets remain the same. More...

Tensor	Transpose (int64_t dim0, int64_t dim1) const
	Transpose a Tensor by swapping dimension `dim0` and `dim1`. More...

Tensor	T () const
	Expects input to be <= 2-D Tensor by swapping dimension 0 and 1. More...

double	Det () const
	Compute the determinant of a 2D square tensor. More...

template<typename T >
T	Item () const

Tensor	Add (const Tensor &value) const
	Adds a tensor and returns the resulting tensor. More...

Tensor	Add (Scalar value) const

Tensor	operator+ (const Tensor &value) const

Tensor	operator+ (Scalar value) const

Tensor	Add_ (const Tensor &value)

Tensor	Add_ (Scalar value)

Tensor	operator+= (const Tensor &value)

Tensor	operator+= (Scalar value)

Tensor	Sub (const Tensor &value) const
	Substracts a tensor and returns the resulting tensor. More...

Tensor	Sub (Scalar value) const

Tensor	operator- (const Tensor &value) const

Tensor	operator- (Scalar value) const

Tensor	Sub_ (const Tensor &value)

Tensor	Sub_ (Scalar value)

Tensor	operator-= (const Tensor &value)

Tensor	operator-= (Scalar value)

Tensor	Mul (const Tensor &value) const
	Multiplies a tensor and returns the resulting tensor. More...

Tensor	Mul (Scalar value) const

Tensor	operator* (const Tensor &value) const

Tensor	operator* (Scalar value) const

Tensor	Mul_ (const Tensor &value)

Tensor	Mul_ (Scalar value)

Tensor	operator*= (const Tensor &value)

Tensor	operator*= (Scalar value)

Tensor	Div (const Tensor &value) const
	Divides a tensor and returns the resulting tensor. More...

Tensor	Div (Scalar value) const

Tensor	operator/ (const Tensor &value) const

Tensor	operator/ (Scalar value) const

Tensor	Div_ (const Tensor &value)

Tensor	Div_ (Scalar value)

Tensor	operator/= (const Tensor &value)

Tensor	operator/= (Scalar value)

Tensor	Sum (const SizeVector &dims, bool keepdim=false) const

Tensor	Mean (const SizeVector &dims, bool keepdim=false) const

Tensor	Prod (const SizeVector &dims, bool keepdim=false) const

Tensor	Min (const SizeVector &dims, bool keepdim=false) const

Tensor	Max (const SizeVector &dims, bool keepdim=false) const

Tensor	ArgMin (const SizeVector &dims) const

Tensor	ArgMax (const SizeVector &dims) const

Tensor	Sqrt () const
	Element-wise square root of a tensor, returns a new tensor. More...

Tensor	Sqrt_ ()
	Element-wise square root of a tensor, in-place. More...

Tensor	Sin () const
	Element-wise sine of a tensor, returning a new tensor. More...

Tensor	Sin_ ()
	Element-wise sine of a tensor, in-place. More...

Tensor	Cos () const
	Element-wise cosine of a tensor, returning a new tensor. More...

Tensor	Cos_ ()
	Element-wise cosine of a tensor, in-place. More...

Tensor	Neg () const
	Element-wise negation of a tensor, returning a new tensor. More...

Tensor	Neg_ ()
	Element-wise negation of a tensor, in-place. More...

Tensor	operator- () const
	Unary minus of a tensor, returning a new tensor. More...

Tensor	Exp () const
	Element-wise exponential of a tensor, returning a new tensor. More...

Tensor	Exp_ ()
	Element-wise base-e exponential of a tensor, in-place. More...

Tensor	Abs () const
	Element-wise absolute value of a tensor, returning a new tensor. More...

Tensor	Abs_ ()
	Element-wise absolute value of a tensor, in-place. More...

Tensor	IsNan () const

Tensor	IsInf () const

Tensor	IsFinite () const

Tensor	Clip (Scalar min_val, Scalar max_val) const

Tensor	Clip_ (Scalar min_val, Scalar max_val)

Tensor	Floor () const
	Element-wise floor value of a tensor, returning a new tensor. More...

Tensor	Ceil () const
	Element-wise ceil value of a tensor, returning a new tensor. More...

Tensor	Round () const
	Element-wise round value of a tensor, returning a new tensor. More...

Tensor	Trunc () const
	Element-wise trunc value of a tensor, returning a new tensor. More...

Tensor	LogicalNot () const

Tensor	LogicalNot_ ()

Tensor	LogicalAnd (const Tensor &value) const

Tensor	operator&& (const Tensor &value) const

Tensor	LogicalAnd (Scalar value) const

Tensor	LogicalAnd_ (const Tensor &value)

Tensor	LogicalAnd_ (Scalar value)

Tensor	LogicalOr (const Tensor &value) const

Tensor	operator\|\| (const Tensor &value) const

Tensor	LogicalOr (Scalar value) const

Tensor	LogicalOr_ (const Tensor &value)

Tensor	LogicalOr_ (Scalar value)

Tensor	LogicalXor (const Tensor &value) const

Tensor	LogicalXor (Scalar value) const

Tensor	LogicalXor_ (const Tensor &value)

Tensor	LogicalXor_ (Scalar value)

Tensor	Gt (const Tensor &value) const
	Element-wise greater-than of tensors, returning a new boolean tensor. More...

Tensor	operator> (const Tensor &value) const

Tensor	Gt (Scalar value) const

Tensor	Gt_ (const Tensor &value)

Tensor	Gt_ (Scalar value)

Tensor	Lt (const Tensor &value) const
	Element-wise less-than of tensors, returning a new boolean tensor. More...

Tensor	operator< (const Tensor &value) const

Tensor	Lt (Scalar value) const

Tensor	Lt_ (const Tensor &value)

Tensor	Lt_ (Scalar value)

Tensor	Ge (const Tensor &value) const

Tensor	operator>= (const Tensor &value) const

Tensor	Ge (Scalar value) const

Tensor	Ge_ (const Tensor &value)

Tensor	Ge_ (Scalar value)

Tensor	Le (const Tensor &value) const

Tensor	operator<= (const Tensor &value) const

Tensor	Le (Scalar value) const

Tensor	Le_ (const Tensor &value)

Tensor	Le_ (Scalar value)

Tensor	Eq (const Tensor &value) const
	Element-wise equals-to of tensors, returning a new boolean tensor. More...

Tensor	operator== (const Tensor &value) const

Tensor	Eq (Scalar value) const

Tensor	Eq_ (const Tensor &value)

Tensor	Eq_ (Scalar value)

Tensor	Ne (const Tensor &value) const
	Element-wise not-equals-to of tensors, returning a new boolean tensor. More...

Tensor	operator!= (const Tensor &value) const

Tensor	Ne (Scalar value) const

Tensor	Ne_ (const Tensor &value)

Tensor	Ne_ (Scalar value)

std::vector< Tensor >	NonZeroNumpy () const

Tensor	NonZero () const

bool	IsNonZero () const

Tensor	All (const utility::optional< SizeVector > &dims=utility::nullopt, bool keepdim=false) const

Tensor	Any (const utility::optional< SizeVector > &dims=utility::nullopt, bool keepdim=false) const

bool	AllEqual (const Tensor &other) const

bool	AllClose (const Tensor &other, double rtol=1e-5, double atol=1e-8) const

Tensor	IsClose (const Tensor &other, double rtol=1e-5, double atol=1e-8) const

bool	IsSame (const Tensor &other) const

template<typename T >
std::vector< T >	ToFlatVector () const
	Retrieve all values as an std::vector, for debugging and testing. More...

bool	IsContiguous () const

Tensor	Contiguous () const

Tensor	Matmul (const Tensor &rhs) const

Tensor	Solve (const Tensor &rhs) const

Tensor	LeastSquares (const Tensor &rhs) const

std::tuple< Tensor, Tensor, Tensor >	LU (const bool permute_l=false) const
	Computes LU factorisation of the 2D square tensor, using A = P * L * U; where P is the permutation matrix, L is the lower-triangular matrix with diagonal elements as 1.0 and U is the upper-triangular matrix, and returns tuple (P, L, U). More...

std::tuple< Tensor, Tensor >	LUIpiv () const
	Computes LU factorisation of the 2D square tensor, using A = P * L * U; where P is the permutation matrix, L is the lower-triangular matrix with diagonal elements as 1.0 and U is the upper-triangular matrix, and returns tuple `output` tensor of shape {n,n} and `ipiv` tensor of shape {n}, where {n,n} is the shape of input tensor. [ipiv, output = open3d.core.lu_ipiv(a)]. More...

Tensor	Triu (const int diagonal=0) const
	Returns the upper triangular matrix of the 2D tensor, above the given diagonal index. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] for a {m,n} shaped tensor. More...

Tensor	Tril (const int diagonal=0) const
	Returns the lower triangular matrix of the 2D tensor, above the given diagonal index. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] where {m, n} is the shape of input tensor. More...

std::tuple< Tensor, Tensor >	Triul (const int diagonal=0) const
	Returns the tuple of upper and lower triangular matrix of the 2D tensor, above and below the given diagonal index. The diagonal elements of lower triangular matrix are taken to be unity. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] where {m, n} is the shape of input tensor. More...

Tensor	Inverse () const

std::tuple< Tensor, Tensor, Tensor >	SVD () const

int64_t	GetLength () const

SizeVector	GetShape () const

const SizeVector &	GetShapeRef () const

int64_t	GetShape (int64_t dim) const

SizeVector	GetStrides () const

const SizeVector &	GetStridesRef () const

int64_t	GetStride (int64_t dim) const

template<typename T >
T *	GetDataPtr ()

template<typename T >
const T *	GetDataPtr () const

void *	GetDataPtr ()

const void *	GetDataPtr () const

Dtype	GetDtype () const

Device	GetDevice () const override

std::shared_ptr< Blob >	GetBlob () const

int64_t	NumElements () const

int64_t	NumDims () const

template<typename T >
void	AssertTemplateDtype () const

DLManagedTensor *	ToDLPack () const
	Convert the Tensor to DLManagedTensor. More...

void	Save (const std::string &file_name) const
	Save tensor to numpy's npy format. More...

Iterator	begin ()

Iterator	end ()

ConstIterator	cbegin () const

ConstIterator	cend () const

ConstIterator	begin () const

ConstIterator	end () const

template<>
	Tensor (const std::vector< bool > &init_vals, const SizeVector &shape, Dtype dtype, const Device &device)

template<>
std::vector< bool >	ToFlatVector () const

template<>
bool	Item () const

Public Member Functions inherited from open3d::core::IsDevice
	IsDevice ()=default

virtual	~IsDevice ()=default

bool	IsCPU () const

bool	IsCUDA () const

Static Public Member Functions
static Tensor	Empty (const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Create a tensor with uninitialized values. More...

static Tensor	EmptyLike (const Tensor &other)

template<typename T >
static Tensor	Full (const SizeVector &shape, T fill_value, Dtype dtype, const Device &device=Device("CPU:0"))
	Create a tensor fill with specified value. More...

static Tensor	Zeros (const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Create a tensor fill with zeros. More...

static Tensor	Ones (const SizeVector &shape, Dtype dtype, const Device &device=Device("CPU:0"))
	Create a tensor fill with ones. More...

template<typename T >
static Tensor	Init (const T val, const Device &device=Device("CPU:0"))

template<typename T >
static Tensor	Init (const std::initializer_list< T > &in_list, const Device &device=Device("CPU:0"))

template<typename T >
static Tensor	Init (const std::initializer_list< std::initializer_list< T >> &in_list, const Device &device=Device("CPU:0"))

template<typename T >
static Tensor	Init (const std::initializer_list< std::initializer_list< std::initializer_list< T >>> &in_list, const Device &device=Device("CPU:0"))

static Tensor	Eye (int64_t n, Dtype dtype, const Device &device)
	Create an identity matrix of size n x n. More...

static Tensor	Diag (const Tensor &input)
	Create a square matrix with specified diagonal elements in input. More...

static Tensor	Arange (const Scalar start, const Scalar stop, const Scalar step=1, const Dtype dtype=core::Int64, const Device &device=core::Device("CPU:0"))
	Create a 1D tensor with evenly spaced values in the given interval. More...

static Tensor	FromDLPack (const DLManagedTensor *dlmt)
	Convert DLManagedTensor to Tensor. More...

static Tensor	Load (const std::string &file_name)
	Load tensor from numpy's npy format. More...

Protected Member Functions
std::string	ScalarPtrToString (const void *ptr) const

Protected Attributes
SizeVector	shape_ = {0}
	SizeVector of the Tensor. shape_[i] is the length of dimension i. More...

SizeVector	strides_ = {1}

void *	data_ptr_ = nullptr

Dtype	dtype_ = core::Undefined
	Data type. More...

std::shared_ptr< Blob >	blob_ = nullptr
	Underlying memory buffer for Tensor. More...

Detailed Description

A Tensor is a "view" of a data Blob with shape, stride, data_ptr. Tensor can also be used to perform numerical operations.

Constructor & Destructor Documentation

◆ Tensor() [1/10]

open3d::core::Tensor::Tensor ( )

inline

◆ Tensor() [2/10]

open3d::core::Tensor::Tensor	(	const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

inline

Constructor for creating a contiguous Tensor.

◆ Tensor() [3/10]

template<typename T >

open3d::core::Tensor::Tensor	(	const std::vector< T > &	init_vals,
		const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

inline

Constructor for creating a contiguous Tensor with initial values.

◆ Tensor() [4/10]

template<typename T >

open3d::core::Tensor::Tensor	(	const T *	init_vals,
		const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

inline

Constructor from raw host buffer. The memory will be copied.

◆ Tensor() [5/10]

open3d::core::Tensor::Tensor	(	const SizeVector &	shape,
		const SizeVector &	strides,
		void *	data_ptr,
		Dtype	dtype,
		const std::shared_ptr< Blob > &	blob
	)

inline

The fully specified constructor. Since you're responsible for creating the Blob, take care of Blob's deleter if the memory is allocated elsewhere. See Blob.h for more details.

◆ Tensor() [6/10]

template<typename T >

open3d::core::Tensor::Tensor	(	std::vector< T > &&	vec,
		const SizeVector &	shape = `{}`
	)

inline

Take ownership of data in std::vector<T>

Create a Tensor from data "moved" from an std::vector<T>. This always produces a Tensor on the CPU device.

Parameters

vec	source for the data as an r-value reference. After the Tensor is created this will not have access to or manage the data any more.
shape	List of dimensions of data in buffer. e.g. `{640, 480, 3}` for a 640x480 RGB image. A 1D {vec.size()} shape is assumed if not specified.

◆ Tensor() [7/10]

open3d::core::Tensor::Tensor	(	void *	data_ptr,
		Dtype	dtype,
		const SizeVector &	shape,
		const SizeVector &	strides = `{}`,
		const Device &	device = `Device("CPU:0")`
	)

inline

Tensor wrapper constructor from raw host buffer.

This creates a Tensor wrapper for externally managed memory. It is the user's responsibility to keep the buffer valid during the lifetime of this Tensor and deallocate it afterwards.

Parameters

data_ptr	Pointer to externally managed buffer.
dtype	Tensor element data type. e.g. `Float32` for single precision float.
shape	List of dimensions of data in buffer. e.g. `{640, 480, 3}` for a 640x480 RGB image.
strides	Number of elements to advance to reach the next element, for every dimension. This will be calculated from the shape assuming a contiguous buffer if not specified. For the above `Float32` image, the value of an element will be read as: image[row, col, ch] = (float ) (data_ptr + sizeof(float) * (row * stride[0] + col * stride[1] + ch * stride[2]));
device	Device containing the data buffer.

◆ Tensor() [8/10]

open3d::core::Tensor::Tensor ( const Tensor & other )

default

Copy constructor performs a "shallow" copy of the Tensor. This takes a lvalue input, e.g. Tensor dst(src).

◆ Tensor() [9/10]

open3d::core::Tensor::Tensor ( Tensor && other )

default

Move constructor performs a "shallow" copy of the Tensor. This takes a rvalue input, e.g. Tensor dst(src[0]).

◆ Tensor() [10/10]

template<>

open3d::core::Tensor::Tensor	(	const std::vector< bool > &	init_vals,
		const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device
	)

inline

Member Function Documentation

◆ Abs()

Tensor open3d::core::Tensor::Abs ( ) const

Element-wise absolute value of a tensor, returning a new tensor.

◆ Abs_()

Tensor open3d::core::Tensor::Abs_ ( )

Element-wise absolute value of a tensor, in-place.

◆ Add() [1/2]

Tensor open3d::core::Tensor::Add ( const Tensor & value ) const

Adds a tensor and returns the resulting tensor.

◆ Add() [2/2]

Tensor open3d::core::Tensor::Add ( Scalar value ) const

◆ Add_() [1/2]

Tensor open3d::core::Tensor::Add_ ( const Tensor & value )

Inplace version of Tensor::Add. Adds a tensor to the current tensor and returns the current tensor.

◆ Add_() [2/2]

Tensor open3d::core::Tensor::Add_ ( Scalar value )

◆ All()

Tensor open3d::core::Tensor::All	(	const utility::optional< SizeVector > &	dims = `utility::nullopt`,
		bool	keepdim = `false`
	)		const

Returns true if all elements in the tensor are true. Only works for boolean tensors.

◆ AllClose()

bool open3d::core::Tensor::AllClose	(	const Tensor &	other,
		double	rtol = `1e-5`,
		double	atol = `1e-8`
	)		const

Returns true if the two tensors are element-wise equal within a tolerance.

If the device is not the same: throws exception.
If the dtype is not the same: throws exception.
If the shape is not the same: returns false.
Returns true if: abs(self - other) <= (atol + rtol * abs(other)).

The equation is not symmetrical, i.e. a.AllClose(b) might not be the same as b.AllClose(a). Also see Numpy's documentation: https://numpy.org/doc/stable/reference/generated/numpy.allclose.html.

TODO: support nan

Parameters

other	The other tensor to compare with.
rtol	Relative tolerance.
atol	Absolute tolerance.

◆ AllEqual()

bool open3d::core::Tensor::AllEqual ( const Tensor & other ) const

Returns true if the two tensors are element-wise equal.

If the device is not the same: throws exception.
If the dtype is not the same: throws exception.
If the shape is not the same: returns false.
Returns true if: the device, dtype and shape are the same and all corresponding elements are equal.

TODO: support nan

Parameters

other The other tensor to compare with.

◆ Any()

Tensor open3d::core::Tensor::Any	(	const utility::optional< SizeVector > &	dims = `utility::nullopt`,
		bool	keepdim = `false`
	)		const

Returns true if any elements in the tensor are true. Only works for boolean tensors.

◆ Append()

Tensor open3d::core::Tensor::Append	(	const Tensor &	other,
		const utility::optional< int64_t > &	axis = `utility::nullopt`
	)		const

Appends the other tensor, along the given axis and returns a copy of the tensor. The other tensors must have same data-type, device, and number of dimensions. All dimensions must be the same, except the dimension along the axis the tensors are to be appended.

This is the same as NumPy's semantics:

https://numpy.org/doc/stable/reference/generated/numpy.append.html

Example:

Tensor a = Tensor::Init<int64_t>({0, 1}, {2, 3});
Tensor b = Tensor::Init<int64_t>({4, 5});
Tensor t1 = a.Append(b, 0);
// t1:
//  [[0 1],
//   [2 3],
//   [4 5]]
//  Tensor[shape={3, 2}, stride={2, 1}, Int64, CPU:0, 0x55555abc6b00]
Tensor t2 = a.Append(b);
// t2:
//  [0 1 2 3 4 5]
//  Tensor[shape={6}, stride={1}, Int64, CPU:0, 0x55555abc6b70]

Parameters

other	Values of this tensor is appended to the tensor.
axis	[optional] The axis along which values are appended. If axis is not given, both tensors are flattened before use.

Returns: A copy of the tensor with values appended to axis. Note that append does not occur in-place: a new array is allocated and filled. If axis is None, out is a flattened tensor.

◆ Arange()

Tensor open3d::core::Tensor::Arange	(	const Scalar	start,
		const Scalar	stop,
		const Scalar	step = `1`,
		const Dtype	dtype = `core::Int64`,
		const Device &	device = `core::Device("CPU:0")`
	)

static

Create a 1D tensor with evenly spaced values in the given interval.

◆ ArgMax()

Tensor open3d::core::Tensor::ArgMax ( const SizeVector & dims ) const

Returns maximum index of the tensor along the given dim. The returned tensor has dtype int64_t, and has the same shape as original tensor except that the reduced dimension is removed.

Parameters

dims	`dims` can only contain a single dimension or all dimensions. If `dims` contains a single dimension, the index is along the specified dimension. If `dims` contains all dimensions, the index is into the flattened tensor.

◆ ArgMin()

Tensor open3d::core::Tensor::ArgMin ( const SizeVector & dims ) const

Returns minimum index of the tensor along the given dim. The returned tensor has dtype int64_t, and has the same shape as original tensor except that the reduced dimension is removed.

Parameters

dims	`dims` can only contain a single dimension or all dimensions. If `dims` contains a single dimension, the index is along the specified dimension. If `dims` contains all dimensions, the index is into the flattened tensor.

◆ AsRvalue() [1/2]

Tensor open3d::core::Tensor::AsRvalue ( )

inline

Convert to rvalue such that the Tensor can be assigned.

E.g. in numpy

tensor_a = tensor_b     # tensor_a is lvalue, tensor_a variable will
                        # now reference tensor_b, that is, tensor_a
                        # and tensor_b share exactly the same memory.
tensor_a[:] = tensor_b  # tensor_a[:] is rvalue, tensor_b's values are
                        # assigned to tensor_a's memory.

◆ AsRvalue() [2/2]

const Tensor open3d::core::Tensor::AsRvalue ( ) const

inline

Convert to constant rvalue.

◆ AssertTemplateDtype()

template<typename T >

void open3d::core::Tensor::AssertTemplateDtype ( ) const

inline

◆ AssignObject()

template<typename Object >

Tensor& open3d::core::Tensor::AssignObject ( const Object & v ) &&

inline

Assign an object to a tensor. The tensor being assigned to must be a scalar tensor of shape {}. The element byte size of the tensor must be the same as the size of the object. The object must be a POD.

◆ AsStrided()

Tensor open3d::core::Tensor::AsStrided	(	const SizeVector &	new_shape,
		const SizeVector &	new_strides
	)		const

Create a Tensor view of specified shape and strides. The underlying buffer and data_ptr offsets remain the same.

◆ begin() [1/2]

Tensor::Iterator open3d::core::Tensor::begin ( )

Returns the beginning of the tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor.

◆ begin() [2/2]

ConstIterator open3d::core::Tensor::begin ( ) const

inline

Returns the beginning of the const tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor. This is equivalent to Tensor::cbegin().

◆ Broadcast()

Tensor open3d::core::Tensor::Broadcast ( const SizeVector & dst_shape ) const

Broadcast Tensor to a new broadcastable shape.

◆ cbegin()

Tensor::ConstIterator open3d::core::Tensor::cbegin ( ) const

Returns the beginning of the const tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor.

◆ Ceil()

Tensor open3d::core::Tensor::Ceil ( ) const

Element-wise ceil value of a tensor, returning a new tensor.

◆ cend()

Tensor::ConstIterator open3d::core::Tensor::cend ( ) const

Returns the end of the const tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor.

◆ Clip()

Tensor open3d::core::Tensor::Clip	(	Scalar	min_val,
		Scalar	max_val
	)		const

Element-wise clipping of tensor values so that resulting values lie in the range [min_val, max_val], returning a new tensor.

Parameters

min_val	Lower bound for output values.
max_val	Upper bound for output values.

◆ Clip_()

Tensor open3d::core::Tensor::Clip_	(	Scalar	min_val,
		Scalar	max_val
	)

Element-wise clipping of tensor values so that resulting values lie in the range [min_val, max_val]. In-place version.

Parameters

min_val	Lower bound for output values.
max_val	Upper bound for output values.

◆ Clone()

Tensor open3d::core::Tensor::Clone ( ) const

inline

Copy Tensor to the same device.

◆ Contiguous()

Tensor open3d::core::Tensor::Contiguous ( ) const

Returns a contiguous Tensor containing the same data in the same device. If self tensor is already contiguous, the same underlying memory will be used.

◆ CopyFrom()

void open3d::core::Tensor::CopyFrom ( const Tensor & other )

Copy Tensor values to current tensor from the source tensor.

◆ Cos()

Tensor open3d::core::Tensor::Cos ( ) const

Element-wise cosine of a tensor, returning a new tensor.

◆ Cos_()

Tensor open3d::core::Tensor::Cos_ ( )

Element-wise cosine of a tensor, in-place.

◆ Det()

double open3d::core::Tensor::Det ( ) const

Compute the determinant of a 2D square tensor.

Returns: returns the determinant of the matrix (double).

◆ Diag()

Tensor open3d::core::Tensor::Diag ( const Tensor & input )

static

Create a square matrix with specified diagonal elements in input.

◆ Div() [1/2]

Tensor open3d::core::Tensor::Div ( const Tensor & value ) const

Divides a tensor and returns the resulting tensor.

◆ Div() [2/2]

Tensor open3d::core::Tensor::Div ( Scalar value ) const

◆ Div_() [1/2]

Tensor open3d::core::Tensor::Div_ ( const Tensor & value )

Inplace version of Tensor::Div. Divides a tensor to the current tensor and returns the current tensor.

◆ Div_() [2/2]

Tensor open3d::core::Tensor::Div_ ( Scalar value )

◆ Empty()

Tensor open3d::core::Tensor::Empty	(	const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

static

Create a tensor with uninitialized values.

◆ EmptyLike()

static Tensor open3d::core::Tensor::EmptyLike ( const Tensor & other )

inlinestatic

Create a tensor with uninitialized values with the same Dtype and Device as the other tensor.

◆ end() [1/2]

Tensor::Iterator open3d::core::Tensor::end ( )

Returns the end of the tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor.

◆ end() [2/2]

ConstIterator open3d::core::Tensor::end ( ) const

inline

Returns the end of the const tensor iterator. The iterator iterates over the first dimension of the tensor. The generated tensor slices share the same memory with the original tensor. This is equivalent to Tensor::cend().

◆ Eq() [1/2]

Tensor open3d::core::Tensor::Eq ( const Tensor & value ) const

Element-wise equals-to of tensors, returning a new boolean tensor.

◆ Eq() [2/2]

Tensor open3d::core::Tensor::Eq ( Scalar value ) const

◆ Eq_() [1/2]

Tensor open3d::core::Tensor::Eq_ ( const Tensor & value )

Element-wise equals-to of tensors, in-place. This operation won't change the tensor's dtype.

◆ Eq_() [2/2]

Tensor open3d::core::Tensor::Eq_ ( Scalar value )

◆ Exp()

Tensor open3d::core::Tensor::Exp ( ) const

Element-wise exponential of a tensor, returning a new tensor.

◆ Exp_()

Tensor open3d::core::Tensor::Exp_ ( )

Element-wise base-e exponential of a tensor, in-place.

◆ Expand()

Tensor open3d::core::Tensor::Expand ( const SizeVector & dst_shape ) const

Expand Tensor to a new broadcastable shape, returning a new view.

Tensors can be expanded to broadcastable shape by setting dimension of size 1 to have stride 0, without allocating new memory.

◆ Eye()

Tensor open3d::core::Tensor::Eye	(	int64_t	n,
		Dtype	dtype,
		const Device &	device
	)

static

Create an identity matrix of size n x n.

◆ Fill()

template<typename S >

void open3d::core::Tensor::Fill ( S v )

inline

Fill the whole Tensor with a scalar value, the scalar will be casted to the Tensor's Dtype.

◆ FillObject()

template<typename Object >

void open3d::core::Tensor::FillObject ( const Object & v )

inline

◆ Flatten()

Tensor open3d::core::Tensor::Flatten	(	int64_t	start_dim = `0`,
		int64_t	end_dim = `-1`
	)		const

Flattens input by reshaping it into a one-dimensional tensor. If start_dim or end_dim are passed, only dimensions starting with start_dim and ending with end_dim are flattened. The order of elements in input is unchanged.

Unlike NumPy’s flatten, which always copies input’s data, this function may return the original object, a view, or copy. If no dimensions are flattened, then the original object input is returned. Otherwise, if input can be viewed as the flattened shape, then that view is returned. Finally, only if the input cannot be viewed as the flattened shape is input’s data copied.

Ref:

https://pytorch.org/docs/stable/tensors.html
aten/src/ATen/native/TensorShape.cpp
aten/src/ATen/TensorUtils.cpp

Parameters

start_dim	The first dimension to flatten (inclusive).
end_dim	The last dimension to flatten, starting from `start_dim` (inclusive).

◆ Floor()

Tensor open3d::core::Tensor::Floor ( ) const

Element-wise floor value of a tensor, returning a new tensor.

◆ FromDLPack()

Tensor open3d::core::Tensor::FromDLPack ( const DLManagedTensor * dlmt )

static

Convert DLManagedTensor to Tensor.

◆ Full()

template<typename T >

static Tensor open3d::core::Tensor::Full	(	const SizeVector &	shape,
		T	fill_value,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

inlinestatic

Create a tensor fill with specified value.

◆ Ge() [1/2]

Tensor open3d::core::Tensor::Ge ( const Tensor & value ) const

Element-wise greater-than-or-equals-to of tensors, returning a new boolean tensor.

◆ Ge() [2/2]

Tensor open3d::core::Tensor::Ge ( Scalar value ) const

◆ Ge_() [1/2]

Tensor open3d::core::Tensor::Ge_ ( const Tensor & value )

Element-wise greater-than-or-equals-to of tensors, in-place. This operation won't change the tensor's dtype.

◆ Ge_() [2/2]

Tensor open3d::core::Tensor::Ge_ ( Scalar value )

◆ GetBlob()

std::shared_ptr<Blob> open3d::core::Tensor::GetBlob ( ) const

inline

◆ GetDataPtr() [1/4]

template<typename T >

T* open3d::core::Tensor::GetDataPtr ( )

inline

◆ GetDataPtr() [2/4]

void* open3d::core::Tensor::GetDataPtr ( )

inline

◆ GetDataPtr() [3/4]

template<typename T >

const T* open3d::core::Tensor::GetDataPtr ( ) const

inline

◆ GetDataPtr() [4/4]

const void* open3d::core::Tensor::GetDataPtr ( ) const

inline

◆ GetDevice()

Device open3d::core::Tensor::GetDevice ( ) const

overridevirtual

Implements open3d::core::IsDevice.

◆ GetDtype()

Dtype open3d::core::Tensor::GetDtype ( ) const

inline

◆ GetItem() [1/2]

Tensor open3d::core::Tensor::GetItem ( const std::vector< TensorKey > & tks ) const

Pythonic getitem for tensor.

Returns a view of the original tensor, if TensorKey only contains TensorKeyMode::Index or TensorKeyMode::Slice. Returns a copy if the TensorKey contains IndexTensor (advanced indexing).

For example, in numpy:

t = np.empty((4, 5), dtype=np.float32)

t1 = t[1, 0:4:2]

The equivalent Open3D C++ calls:

Tensor t({4, 5}, core::Float32);

Tensor t1 = t.GetItem({TensorIndex(2), TensorSlice(0, 4, 2)});

open3d::core::Float32

const Dtype Float32

Definition: Dtype.cpp:42

◆ GetItem() [2/2]

Tensor open3d::core::Tensor::GetItem ( const TensorKey & tk ) const

Pythonic getitem for tensor.

Returns a view of the original tensor, if TensorKey is TensorKeyMode::Index or TensorKeyMode::Slice. Returns a copy if the TensorKey contains TensorKeyMode::IndexTensor (advanced indexing).

For example, in numpy:

t = np.empty((4, 5), dtype=np.float32)
t1 = t[2]
t2 = t[0:4:2]

The equivalent Open3D C++ calls:

Tensor t({4, 5}, core::Float32);
Tensor t1 = t.GetItem(TensorIndex(2));
Tensor t2 = t.GetItem(TensorSlice(0, 4, 2));

◆ GetLength()

int64_t open3d::core::Tensor::GetLength ( ) const

inline

Returns the size of the first dimension. If NumDims() == 0, an exception will be thrown.

◆ GetShape() [1/2]

SizeVector open3d::core::Tensor::GetShape ( ) const

inline

◆ GetShape() [2/2]

int64_t open3d::core::Tensor::GetShape ( int64_t dim ) const

inline

◆ GetShapeRef()

const SizeVector& open3d::core::Tensor::GetShapeRef ( ) const

inline

◆ GetStride()

int64_t open3d::core::Tensor::GetStride ( int64_t dim ) const

inline

◆ GetStrides()

SizeVector open3d::core::Tensor::GetStrides ( ) const

inline

◆ GetStridesRef()

const SizeVector& open3d::core::Tensor::GetStridesRef ( ) const

inline

◆ Gt() [1/2]

Tensor open3d::core::Tensor::Gt ( const Tensor & value ) const

Element-wise greater-than of tensors, returning a new boolean tensor.

◆ Gt() [2/2]

Tensor open3d::core::Tensor::Gt ( Scalar value ) const

◆ Gt_() [1/2]

Tensor open3d::core::Tensor::Gt_ ( const Tensor & value )

Element-wise greater-than of tensors, in-place. This operation won't change the tensor's dtype.

◆ Gt_() [2/2]

Tensor open3d::core::Tensor::Gt_ ( Scalar value )

◆ IndexExtract()

Tensor open3d::core::Tensor::IndexExtract	(	int64_t	dim,
		int64_t	idx
	)		const

Extract the idx -th sub-tensor in dimension dim. After IndexExtract, the dimension dim will be removed.

◆ IndexGet()

Tensor open3d::core::Tensor::IndexGet ( const std::vector< Tensor > & index_tensors ) const

Advanced indexing getter. This will always allocate a new Tensor.

We use the Numpy advanced indexing semantics, see: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html

◆ IndexSet()

void open3d::core::Tensor::IndexSet	(	const std::vector< Tensor > &	index_tensors,
		const Tensor &	src_tensor
	)

Advanced indexing getter.

We use the Numpy advanced indexing semantics, see: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html

Note: Only support 1D index tensors. Note: Only support advanced indices are all next to each other.

◆ Init() [1/4]

template<typename T >

static Tensor open3d::core::Tensor::Init	(	const std::initializer_list< std::initializer_list< std::initializer_list< T >>> &	in_list,
		const Device &	device = `Device("CPU:0")`
	)

inlinestatic

Create a 3-D tensor with nested initializer list, e.g., core::Tensor::Init<float>({{{0, 1}, {2, 3}}, {{4, 5}, {6, 7}}});

◆ Init() [2/4]

template<typename T >

static Tensor open3d::core::Tensor::Init	(	const std::initializer_list< std::initializer_list< T >> &	in_list,
		const Device &	device = `Device("CPU:0")`
	)

inlinestatic

Create a 2-D tensor with nested initializer list, e.g., core::Tensor::Init<float>({{0, 1, 2}, {3, 4, 5}});

◆ Init() [3/4]

template<typename T >

static Tensor open3d::core::Tensor::Init	(	const std::initializer_list< T > &	in_list,
		const Device &	device = `Device("CPU:0")`
	)

inlinestatic

Create a 1-D tensor with initializer list, e.g., core::Tensor::Init<float>({0, 1, 2});

◆ Init() [4/4]

template<typename T >

static Tensor open3d::core::Tensor::Init	(	const T	val,
		const Device &	device = `Device("CPU:0")`
	)

inlinestatic

Create a 0-D tensor (scalar) with given value, e.g., core::Tensor::Init<float>(0);

◆ Inverse()

Tensor open3d::core::Tensor::Inverse ( ) const

Computes the matrix inversion of the square matrix *this with LU factorization and returns the result.

◆ IsClose()

Tensor open3d::core::Tensor::IsClose	(	const Tensor &	other,
		double	rtol = `1e-5`,
		double	atol = `1e-8`
	)		const

Element-wise version of Tensor::AllClose().

If the device is not the same: throws exception.
If the dtype is not the same: throws exception.
If the shape is not the same: throws exception.
For each element in the returned tensor: abs(self - other) <= (atol + rtol * abs(other)).

The equation is not symmetrial, i.e. a.AllClose(b) might not be the same as b.AllClose(a). Also see Numpy's documentation: https://numpy.org/doc/stable/reference/generated/numpy.allclose.html.

TODO: support nan

Parameters

other	The other tensor to compare with.
rtol	Relative tolerance.
atol	Absolute tolerance.

Returns: A boolean tensor indicating where the tensor is close.

◆ IsContiguous()

bool open3d::core::Tensor::IsContiguous ( ) const

inline

Returns True if the underlying memory buffer is contiguous. A contiguous Tensor's data_ptr_ does not need to point to the beginning of blob_.

◆ IsFinite()

Tensor open3d::core::Tensor::IsFinite ( ) const

Element-wise check for finite values (not Inf or NaN) in a tensor, returning a new Boolean tensor. Non-floating point tensors return all True values.

◆ IsInf()

Tensor open3d::core::Tensor::IsInf ( ) const

Element-wise check for Infinity values in a tensor, returning a new Boolean tensor. Non-floating point tensors return all False values.

◆ IsNan()

Tensor open3d::core::Tensor::IsNan ( ) const

Element-wise check for NaN values in a tensor, returning a new Boolean tensor. Non-floating point tensors return all False values.

◆ IsNonZero()

bool open3d::core::Tensor::IsNonZero ( ) const

Evaluate a single-element Tensor as a boolean value. This can be used to implement Tensor.__bool__() in Python, e.g.

assert Tensor([True])         # Passes.
assert Tensor([123])          # Passes.
assert Tensor([False])        # AssertionError.
assert Tensor([0])            # AssertionError.
assert Tensor([True, False])  # ValueError: cannot be evaluated as bool.

◆ IsSame()

bool open3d::core::Tensor::IsSame ( const Tensor & other ) const

Returns true iff the tensor is the other tensor. This means that, the two tensors have the same underlying memory, device, dtype, shape, strides and etc.

◆ Item() [1/2]

template<typename T >

T open3d::core::Tensor::Item ( ) const

inline

Helper function to return scalar value of a scalar Tensor, the Tensor must have empty shape.

◆ Item() [2/2]

template<>

bool open3d::core::Tensor::Item ( ) const

inline

◆ Le() [1/2]

Tensor open3d::core::Tensor::Le ( const Tensor & value ) const

Element-wise less-than-or-equals-to of tensors, returning a new boolean tensor.

◆ Le() [2/2]

Tensor open3d::core::Tensor::Le ( Scalar value ) const

◆ Le_() [1/2]

Tensor open3d::core::Tensor::Le_ ( const Tensor & value )

Element-wise less-than-or-equals-to of tensors, in-place. This operation won't change the tensor's dtype.

◆ Le_() [2/2]

Tensor open3d::core::Tensor::Le_ ( Scalar value )

◆ LeastSquares()

Tensor open3d::core::Tensor::LeastSquares ( const Tensor & rhs ) const

Solves the linear system AX = B with QR decomposition and returns X. A is a (m, n) matrix with m >= n.

◆ Load()

Tensor open3d::core::Tensor::Load ( const std::string & file_name )

static

Load tensor from numpy's npy format.

◆ LogicalAnd() [1/2]

Tensor open3d::core::Tensor::LogicalAnd ( const Tensor & value ) const

Element-wise logical and of a tensor, returning a new boolean tensor.

If the tensor is not boolean, zero will be treated as False, while non-zero values will be treated as True.

◆ LogicalAnd() [2/2]

Tensor open3d::core::Tensor::LogicalAnd ( Scalar value ) const

◆ LogicalAnd_() [1/2]

Tensor open3d::core::Tensor::LogicalAnd_ ( const Tensor & value )

Element-wise logical and of tensors, in-place. This operation won't change the tensor's dtype.

If the tensor is not boolean, 0 will be treated as False, while non-zero will be treated as True. The tensor will be filled with 0 or 1 casted to the tensor's dtype.

◆ LogicalAnd_() [2/2]

Tensor open3d::core::Tensor::LogicalAnd_ ( Scalar value )

◆ LogicalNot()

Tensor open3d::core::Tensor::LogicalNot ( ) const

Element-wise logical not of a tensor, returning a new boolean tensor.

If the tensor is not boolean, 0 will be treated as False, while non-zero will be treated as True.

◆ LogicalNot_()

Tensor open3d::core::Tensor::LogicalNot_ ( )

Element-wise logical not of a tensor, in-place. This operation won't change the tensor's dtype.

If the tensor is not boolean, 0 will be treated as False, while non-zero will be treated as True. The tensor will be filled with 0 or 1 casted to the tensor's dtype.

◆ LogicalOr() [1/2]

Tensor open3d::core::Tensor::LogicalOr ( const Tensor & value ) const

Element-wise logical or of tensors, returning a new boolean tensor.

If the tensor is not boolean, zero will be treated as False, while non-zero values will be treated as True.

◆ LogicalOr() [2/2]

Tensor open3d::core::Tensor::LogicalOr ( Scalar value ) const

◆ LogicalOr_() [1/2]

Tensor open3d::core::Tensor::LogicalOr_ ( const Tensor & value )

Element-wise logical or of tensors, in-place. This operation won't change the tensor's dtype.

If the tensor is not boolean, 0 will be treated as False, while non-zero will be treated as True. The tensor will be filled with 0 or 1 casted to the tensor's dtype.

◆ LogicalOr_() [2/2]

Tensor open3d::core::Tensor::LogicalOr_ ( Scalar value )

◆ LogicalXor() [1/2]

Tensor open3d::core::Tensor::LogicalXor ( const Tensor & value ) const

Element-wise logical exclusive-or of tensors, returning a new boolean tensor.

If the tensor is not boolean, zero will be treated as False, while non-zero values will be treated as True.

◆ LogicalXor() [2/2]

Tensor open3d::core::Tensor::LogicalXor ( Scalar value ) const

◆ LogicalXor_() [1/2]

Tensor open3d::core::Tensor::LogicalXor_ ( const Tensor & value )

Element-wise logical exclusive-or of tensors, in-place. This operation won't change the tensor's dtype.

If the tensor is not boolean, zero will be treated as False, while non-zero values will be treated as True. The tensor will be filled with 0 or 1 casted to the tensor's dtype.

◆ LogicalXor_() [2/2]

Tensor open3d::core::Tensor::LogicalXor_ ( Scalar value )

◆ Lt() [1/2]

Tensor open3d::core::Tensor::Lt ( const Tensor & value ) const

Element-wise less-than of tensors, returning a new boolean tensor.

◆ Lt() [2/2]

Tensor open3d::core::Tensor::Lt ( Scalar value ) const

◆ Lt_() [1/2]

Tensor open3d::core::Tensor::Lt_ ( const Tensor & value )

Element-wise less-than of tensors, in-place. This operation won't change the tensor's dtype.

◆ Lt_() [2/2]

Tensor open3d::core::Tensor::Lt_ ( Scalar value )

◆ LU()

std::tuple< Tensor, Tensor, Tensor > open3d::core::Tensor::LU ( const bool permute_l = false ) const

Computes LU factorisation of the 2D square tensor, using A = P * L * U; where P is the permutation matrix, L is the lower-triangular matrix with diagonal elements as 1.0 and U is the upper-triangular matrix, and returns tuple (P, L, U).

Parameters

permute_l [optional input] If true: returns L as P * L.

Returns: Tuple (P, L, U).

◆ LUIpiv()

std::tuple< Tensor, Tensor > open3d::core::Tensor::LUIpiv ( ) const

Computes LU factorisation of the 2D square tensor, using A = P * L * U; where P is the permutation matrix, L is the lower-triangular matrix with diagonal elements as 1.0 and U is the upper-triangular matrix, and returns tuple output tensor of shape {n,n} and ipiv tensor of shape {n}, where {n,n} is the shape of input tensor. [ipiv, output = open3d.core.lu_ipiv(a)].

Returns: Tuple {ipiv, output}. Where ipiv is a 1D integer pivort indices tensor. It contains the pivot indices, indicating row i of the matrix was interchanged with row ipiv(i)); and output it has L as lower triangular values and U as upper triangle values including the main diagonal (diagonal elements of L to be taken as unity).

◆ Matmul()

Tensor open3d::core::Tensor::Matmul ( const Tensor & rhs ) const

Computes matrix multiplication with *this and rhs and returns the result.

◆ Max()

Tensor open3d::core::Tensor::Max	(	const SizeVector &	dims,
		bool	keepdim = `false`
	)		const

Returns max of the tensor along the given dims.

Parameters

dims	A list of dimensions to be reduced.
keepdim	If true, the reduced dims will be retained as size 1.

◆ Mean()

Tensor open3d::core::Tensor::Mean	(	const SizeVector &	dims,
		bool	keepdim = `false`
	)		const

Returns the mean of the tensor along the given dims.

Parameters

dims	A list of dimensions to be reduced.
keepdim	If true, the reduced dims will be retained as size 1.

◆ Min()

Tensor open3d::core::Tensor::Min	(	const SizeVector &	dims,
		bool	keepdim = `false`
	)		const

Returns min of the tensor along the given dims.

Parameters

dims	A list of dimensions to be reduced.
keepdim	If true, the reduced dims will be retained as size 1.

◆ Mul() [1/2]

Tensor open3d::core::Tensor::Mul ( const Tensor & value ) const

Multiplies a tensor and returns the resulting tensor.

◆ Mul() [2/2]

Tensor open3d::core::Tensor::Mul ( Scalar value ) const

◆ Mul_() [1/2]

Tensor open3d::core::Tensor::Mul_ ( const Tensor & value )

Inplace version of Tensor::Mul. Multiplies a tensor to the current tensor and returns the current tensor.

◆ Mul_() [2/2]

Tensor open3d::core::Tensor::Mul_ ( Scalar value )

◆ Ne() [1/2]

Tensor open3d::core::Tensor::Ne ( const Tensor & value ) const

Element-wise not-equals-to of tensors, returning a new boolean tensor.

◆ Ne() [2/2]

Tensor open3d::core::Tensor::Ne ( Scalar value ) const

◆ Ne_() [1/2]

Tensor open3d::core::Tensor::Ne_ ( const Tensor & value )

Element-wise equals-to of tensors, in-place. This operation won't change the tensor's dtype.

◆ Ne_() [2/2]

Tensor open3d::core::Tensor::Ne_ ( Scalar value )

◆ Neg()

Tensor open3d::core::Tensor::Neg ( ) const

Element-wise negation of a tensor, returning a new tensor.

◆ Neg_()

Tensor open3d::core::Tensor::Neg_ ( )

Element-wise negation of a tensor, in-place.

◆ NonZero()

Tensor open3d::core::Tensor::NonZero ( ) const

Find the indices of the elements that are non-zero. Returns an int64 tensor of shape {num_dims, num_non_zeros}, where the i-th row contains the indices of the non-zero elements in i-th dimension of the original tensor.

◆ NonZeroNumpy()

std::vector< Tensor > open3d::core::Tensor::NonZeroNumpy ( ) const

Find the indices of the elements that are non-zero. Returns a vector of int64 Tensors, each containing the indices of the non-zero elements in each dimension.

◆ NumDims()

int64_t open3d::core::Tensor::NumDims ( ) const

inline

◆ NumElements()

int64_t open3d::core::Tensor::NumElements ( ) const

inline

◆ Ones()

Tensor open3d::core::Tensor::Ones	(	const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

static

Create a tensor fill with ones.

◆ operator!=()

Tensor open3d::core::Tensor::operator!= ( const Tensor & value ) const

inline

◆ operator&&()

Tensor open3d::core::Tensor::operator&& ( const Tensor & value ) const

inline

◆ operator*() [1/2]

Tensor open3d::core::Tensor::operator* ( const Tensor & value ) const

inline

◆ operator*() [2/2]

Tensor open3d::core::Tensor::operator* ( Scalar value ) const

inline

◆ operator*=() [1/2]

Tensor open3d::core::Tensor::operator*= ( const Tensor & value )

inline

◆ operator*=() [2/2]

Tensor open3d::core::Tensor::operator*= ( Scalar value )

inline

◆ operator+() [1/2]

Tensor open3d::core::Tensor::operator+ ( const Tensor & value ) const

inline

◆ operator+() [2/2]

Tensor open3d::core::Tensor::operator+ ( Scalar value ) const

inline

◆ operator+=() [1/2]

Tensor open3d::core::Tensor::operator+= ( const Tensor & value )

inline

◆ operator+=() [2/2]

Tensor open3d::core::Tensor::operator+= ( Scalar value )

inline

◆ operator-() [1/3]

Tensor open3d::core::Tensor::operator- ( ) const

inline

Unary minus of a tensor, returning a new tensor.

◆ operator-() [2/3]

Tensor open3d::core::Tensor::operator- ( const Tensor & value ) const

inline

◆ operator-() [3/3]

Tensor open3d::core::Tensor::operator- ( Scalar value ) const

inline

◆ operator-=() [1/2]

Tensor open3d::core::Tensor::operator-= ( const Tensor & value )

inline

◆ operator-=() [2/2]

Tensor open3d::core::Tensor::operator-= ( Scalar value )

inline

◆ operator/() [1/2]

Tensor open3d::core::Tensor::operator/ ( const Tensor & value ) const

inline

◆ operator/() [2/2]

Tensor open3d::core::Tensor::operator/ ( Scalar value ) const

inline

◆ operator/=() [1/2]

Tensor open3d::core::Tensor::operator/= ( const Tensor & value )

inline

◆ operator/=() [2/2]

Tensor open3d::core::Tensor::operator/= ( Scalar value )

inline

◆ operator<()

Tensor open3d::core::Tensor::operator< ( const Tensor & value ) const

inline

◆ operator<=()

Tensor open3d::core::Tensor::operator<= ( const Tensor & value ) const

inline

◆ operator=() [1/5]

template<typename T >

Tensor& open3d::core::Tensor::operator= ( const T v ) &&

inline

Tensor assignment rvalue = scalar, e.g. tensor_a[0] = 100 Implicit casting is performed to the underlying dtype.

Note that we don't have lvalue = scalar, e.g. we don't support Tensor a_slice = tensor_a[0]; a_slice = 100;

◆ operator=() [2/5]

Tensor & open3d::core::Tensor::operator= ( const Tensor & other ) &

Tensor assignment lvalue = lvalue, e.g. tensor_a = tensor_b. This results in a "shallow" copy.

◆ operator=() [3/5]

Tensor & open3d::core::Tensor::operator= ( const Tensor & other ) &&

Tensor assignment rvalue = lvalue, e.g. tensor_a[0] = tensor_b.

Tensor assignment rvalue = lvalue, e.g. tensor_a[0] = tensor_b. An actual copy of the data will be performed.

◆ operator=() [4/5]

Tensor & open3d::core::Tensor::operator= ( Tensor && other ) &

Tensor assignment lvalue = rvalue, e.g. tensor_a = tensor_b[0]. This results in a "shallow" copy.

◆ operator=() [5/5]

Tensor & open3d::core::Tensor::operator= ( Tensor && other ) &&

Tensor assignment rvalue = rvalue, e.g. tensor_a[0] = tensor_b[0].

Tensor assignment rvalue = rvalue, e.g. tensor_a[0] = tensor_b[0]. An actual copy of the data will be performed.

◆ operator==()

Tensor open3d::core::Tensor::operator== ( const Tensor & value ) const

inline

◆ operator>()

Tensor open3d::core::Tensor::operator> ( const Tensor & value ) const

inline

◆ operator>=()

Tensor open3d::core::Tensor::operator>= ( const Tensor & value ) const

inline

◆ operator[]()

Tensor open3d::core::Tensor::operator[] ( int64_t i ) const

Extract the i-th Tensor along the first axis, returning a new view.

◆ operator||()

Tensor open3d::core::Tensor::operator|| ( const Tensor & value ) const

inline

◆ Permute()

Tensor open3d::core::Tensor::Permute ( const SizeVector & dims ) const

Permute (dimension shuffle) the Tensor, returns a view.

Parameters

dims	The desired ordering of dimensions.

Returns: A Tensor with the desired ordering of the dimensions.

◆ Prod()

Tensor open3d::core::Tensor::Prod	(	const SizeVector &	dims,
		bool	keepdim = `false`
	)		const

Returns the product of the tensor along the given dims.

Parameters

dims	A list of dimensions to be reduced.
keepdim	If true, the reduced dims will be retained as size 1.

◆ ReinterpretCast()

Tensor open3d::core::Tensor::ReinterpretCast ( const core::Dtype & dtype ) const

Tensor reinterpret cast operator. It changes the tensor's dtype without changing the underlying memory blob itself. The byte-size of dtype must be same as the original dtype before casting.

◆ Reshape()

Tensor open3d::core::Tensor::Reshape ( const SizeVector & dst_shape ) const

Returns a tensor with the same data and number of elements as input, but with the specified shape. When possible, the returned tensor will be a view of input. Otherwise, it will be a copy.

Contiguous inputs and inputs with compatible strides can be reshaped without copying, but you should not depend on the copying vs. viewing behavior.

Ref:

https://pytorch.org/docs/stable/tensors.html
aten/src/ATen/native/TensorShape.cpp
aten/src/ATen/TensorUtils.cpp

◆ Reverse()

Tensor open3d::core::Tensor::Reverse ( ) const

Reverse a Tensor's elements by viewing the tensor as a 1D array.

◆ Round()

Tensor open3d::core::Tensor::Round ( ) const

Element-wise round value of a tensor, returning a new tensor.

◆ Save()

void open3d::core::Tensor::Save ( const std::string & file_name ) const

Save tensor to numpy's npy format.

◆ ScalarPtrToString()

std::string open3d::core::Tensor::ScalarPtrToString ( const void * ptr ) const

protected

◆ SetItem() [1/3]

Tensor open3d::core::Tensor::SetItem	(	const std::vector< TensorKey > &	tks,
		const Tensor &	value
	)

Pythonic setitem for tensor.

For example, in numpy:

t = np.empty((4, 5), dtype=np.float32)

t[2, 0:4:2] = np.empty((2, 5), dtype=np.float32)

The equivalent Open3D C++ calls:

Tensor t({4, 5}, core::Float32);
t.SetItem({TensorIndex(2), TensorSlice(0, 4, 2)},
          Tensor({2, 5}, core::Float32));

◆ SetItem() [2/3]

Tensor open3d::core::Tensor::SetItem ( const Tensor & value )

Set all items. Equivalent to tensor[:] = value in Python.

◆ SetItem() [3/3]

Tensor open3d::core::Tensor::SetItem	(	const TensorKey &	tk,
		const Tensor &	value
	)

Pythonic setitem for tensor.

For example, in numpy:

t = np.empty((4, 5), dtype=np.float32)
t[2] = np.empty((5,), dtype=np.float32)
t[0:4:2] = np.empty((2, 5), dtype=np.float32)

The equivalent Open3D C++ calls:

Tensor t({4, 5}, core::Float32);
t.SetItem(TensorIndex(2), Tensor({5}, core::Float32));
t.SetItem(TensorSlice(0, 4, 2), Tensor({2, 5}, core::Float32));

◆ Sin()

Tensor open3d::core::Tensor::Sin ( ) const

Element-wise sine of a tensor, returning a new tensor.

◆ Sin_()

Tensor open3d::core::Tensor::Sin_ ( )

Element-wise sine of a tensor, in-place.

◆ Slice()

Tensor open3d::core::Tensor::Slice	(	int64_t	dim,
		int64_t	start,
		int64_t	stop,
		int64_t	step = `1`
	)		const

Slice Tensor.

Parameters

dim	The dimension to slice.
start	The start index (inclusive).
stop	The end index (exclusive).
step	Pick one element for every `step` elements.

◆ Solve()

Tensor open3d::core::Tensor::Solve ( const Tensor & rhs ) const

Solves the linear system AX = B with LU decomposition and returns X. A must be a square matrix.

◆ Sqrt()

Tensor open3d::core::Tensor::Sqrt ( ) const

Element-wise square root of a tensor, returns a new tensor.

◆ Sqrt_()

Tensor open3d::core::Tensor::Sqrt_ ( )

Element-wise square root of a tensor, in-place.

◆ Sub() [1/2]

Tensor open3d::core::Tensor::Sub ( const Tensor & value ) const

Substracts a tensor and returns the resulting tensor.

◆ Sub() [2/2]

Tensor open3d::core::Tensor::Sub ( Scalar value ) const

◆ Sub_() [1/2]

Tensor open3d::core::Tensor::Sub_ ( const Tensor & value )

Inplace version of Tensor::Sub. Substracts a tensor to the current tensor and returns the current tensor.

◆ Sub_() [2/2]

Tensor open3d::core::Tensor::Sub_ ( Scalar value )

◆ Sum()

Tensor open3d::core::Tensor::Sum	(	const SizeVector &	dims,
		bool	keepdim = `false`
	)		const

Returns the sum of the tensor along the given dims.

Parameters

dims	A list of dimensions to be reduced.
keepdim	If true, the reduced dims will be retained as size 1.

◆ SVD()

std::tuple< Tensor, Tensor, Tensor > open3d::core::Tensor::SVD ( ) const

Computes the matrix SVD decomposition A = U S VT and returns the result. Note VT (V transpose) is returned instead of V.

◆ T()

Tensor open3d::core::Tensor::T ( ) const

Expects input to be <= 2-D Tensor by swapping dimension 0 and 1.

0-D and 1-D Tensor remains the same.

◆ To() [1/3]

Tensor open3d::core::Tensor::To	(	const Device &	device,
		bool	copy = `false`
	)		const

Returns a tensor with the specified device.

Parameters

device	The targeted device to convert to.
copy	If true, a new tensor is always created; if false, the copy is avoided when the original tensor is already on the targeted device.

◆ To() [2/3]

Tensor open3d::core::Tensor::To	(	const Device &	device,
		Dtype	dtype,
		bool	copy = `false`
	)		const

Returns a tensor with the specified device and dtype.

Parameters

device	The targeted device to convert to.
dtype	The targeted dtype to convert to.
copy	If true, a new tensor is always created; if false, the copy is avoided when the original tensor is already on the targeted device and have the targeted dtype.

◆ To() [3/3]

Tensor open3d::core::Tensor::To	(	Dtype	dtype,
		bool	copy = `false`
	)		const

Returns a tensor with the specified dtype.

Parameters

dtype	The targeted dtype to convert to.
copy	If true, a new tensor is always created; if false, the copy is avoided when the original tensor already has the targeted dtype.

◆ ToDLPack()

DLManagedTensor * open3d::core::Tensor::ToDLPack ( ) const

Convert the Tensor to DLManagedTensor.

◆ ToFlatVector() [1/2]

template<typename T >

std::vector<T> open3d::core::Tensor::ToFlatVector ( ) const

inline

Retrieve all values as an std::vector, for debugging and testing.

◆ ToFlatVector() [2/2]

template<>

std::vector<bool> open3d::core::Tensor::ToFlatVector ( ) const

inline

◆ ToString()

std::string open3d::core::Tensor::ToString	(	bool	with_suffix = `true`,
		const std::string &	indent = `""`
	)		const

◆ Transpose()

Tensor open3d::core::Tensor::Transpose	(	int64_t	dim0,
		int64_t	dim1
	)		const

Transpose a Tensor by swapping dimension dim0 and dim1.

Parameters

dim0	The first dimension to be transposed.
dim1	The second dimension to be transposed.

◆ Tril()

Tensor open3d::core::Tensor::Tril ( const int diagonal = 0 ) const

Returns the lower triangular matrix of the 2D tensor, above the given diagonal index. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] where {m, n} is the shape of input tensor.

Parameters

diagonal value of [col - row], below which the elements are to be taken for lower triangular matrix.

◆ Triu()

Tensor open3d::core::Tensor::Triu ( const int diagonal = 0 ) const

Returns the upper triangular matrix of the 2D tensor, above the given diagonal index. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] for a {m,n} shaped tensor.

Parameters

diagonal value of [col - row], above which the elements are to be taken for upper triangular matrix.

◆ Triul()

std::tuple< Tensor, Tensor > open3d::core::Tensor::Triul ( const int diagonal = 0 ) const

Returns the tuple of upper and lower triangular matrix of the 2D tensor, above and below the given diagonal index. The diagonal elements of lower triangular matrix are taken to be unity. [The value of diagonal = col - row, therefore 0 is the main diagonal (row = col), and it shifts towards right for positive values (for diagonal = 1, col - row = 1), and towards left for negative values. The value of the diagonal parameter must be between [-m, n] where {m, n} is the shape of input tensor.

Parameters

diagonal value of [col - row], above and below which the elements are to be taken for upper (diag. included) and lower triangular matrix.

◆ Trunc()

Tensor open3d::core::Tensor::Trunc ( ) const

Element-wise trunc value of a tensor, returning a new tensor.

◆ View()

Tensor open3d::core::Tensor::View ( const SizeVector & dst_shape ) const

Returns a new tensor view with the same data but of a different shape.

The returned tensor shares the same data and must have the same number of elements, but may have a different size. For a tensor to be viewed, the new view size must be compatible with its original size and stride, i.e., each new view dimension must either be a subspace of an original dimension, or only span across original dimensions d, d+1, ..., d+kd,d+1, ..., d+k that satisfy the following contiguity-like condition that for all i = 0, ..., k-1, strides[i] = stride[i + 1] * shape[i + 1].

Otherwise, Contiguous() needs to be called before the tensor can be viewed. See also: Reshape(), which returns a view if the shapes are compatible, and copies (equivalent to calling Contiguous()) otherwise.

Ref:

https://pytorch.org/docs/stable/tensors.html
aten/src/ATen/native/TensorShape.cpp
aten/src/ATen/TensorUtils.cpp

◆ Zeros()

Tensor open3d::core::Tensor::Zeros	(	const SizeVector &	shape,
		Dtype	dtype,
		const Device &	device = `Device("CPU:0")`
	)

static

Create a tensor fill with zeros.

Field Documentation

◆ blob_

std::shared_ptr<Blob> open3d::core::Tensor::blob_ = nullptr

protected

Underlying memory buffer for Tensor.

◆ data_ptr_

void* open3d::core::Tensor::data_ptr_ = nullptr

protected

Data pointer pointing to the beginning element of the Tensor.

Note that this is not necessarily the same as blob_.GetDataPtr(). When this happens, it means that the beginning element of the Tensor is not located a the beginning of the underlying blob. This could happen, for instance, at slicing:

// a.GetDataPtr() == a.GetBlob().GetDataPtr()
Tensor a({2, 3}, dtype, "CPU:0");
// b.GetDataPtr() != b.GetBlob().GetDataPtr()
b = a[1];

◆ dtype_

Dtype open3d::core::Tensor::dtype_ = core::Undefined

protected

Data type.

◆ shape_

SizeVector open3d::core::Tensor::shape_ = {0}

protected

SizeVector of the Tensor. shape_[i] is the length of dimension i.

◆ strides_

SizeVector open3d::core::Tensor::strides_ = {1}

protected

Stride of a Tensor. The stride of a n-dimensional tensor is also n-dimensional. Stride(i) is the number of elements (not bytes) to jump in a continuous memory space before reaching the next element in dimension i. For example, a 2x3x4 float32 dense tensor has shape(2, 3, 4) and stride(12, 4, 1). A slicing operation performed on the tensor can change the shape and stride.

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

/home/runner/work/Open3D/Open3D/cpp/open3d/core/Tensor.h (9238339 (Tue Mar 14 18:49:09 2023 -0700))
/home/runner/work/Open3D/Open3D/cpp/open3d/core/Tensor.cpp (9238339 (Tue Mar 14 18:49:09 2023 -0700))

Data Structures

Public Member Functions

Static Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ Tensor() [1/10]

◆ Tensor() [2/10]

◆ Tensor() [3/10]

◆ Tensor() [4/10]

◆ Tensor() [5/10]

◆ Tensor() [6/10]

◆ Tensor() [7/10]

◆ Tensor() [8/10]

◆ Tensor() [9/10]

◆ Tensor() [10/10]

Member Function Documentation

◆ Abs()

◆ Abs_()

◆ Add() [1/2]

◆ Add() [2/2]

◆ Add_() [1/2]

◆ Add_() [2/2]

◆ All()

◆ AllClose()

◆ AllEqual()

◆ Any()

◆ Append()

◆ Arange()

◆ ArgMax()

◆ ArgMin()

◆ AsRvalue() [1/2]

◆ AsRvalue() [2/2]

◆ AssertTemplateDtype()

◆ AssignObject()

◆ AsStrided()

◆ begin() [1/2]

◆ begin() [2/2]

◆ Broadcast()

◆ cbegin()

◆ Ceil()

◆ cend()

◆ Clip()

◆ Clip_()

◆ Clone()

◆ Contiguous()

◆ CopyFrom()

◆ Cos()

◆ Cos_()

◆ Det()

◆ Diag()

◆ Div() [1/2]

◆ Div() [2/2]

◆ Div_() [1/2]

◆ Div_() [2/2]

◆ Empty()

◆ EmptyLike()

◆ end() [1/2]

◆ end() [2/2]

◆ Eq() [1/2]

◆ Eq() [2/2]

◆ Eq_() [1/2]

◆ Eq_() [2/2]

◆ Exp()

◆ Exp_()

◆ Expand()

◆ Eye()

◆ Fill()

◆ FillObject()

◆ Flatten()

◆ Floor()

◆ FromDLPack()

◆ Full()

◆ Ge() [1/2]

◆ Ge() [2/2]

◆ Ge_() [1/2]

◆ Ge_() [2/2]

◆ GetBlob()

◆ GetDataPtr() [1/4]