open3d::core::shape_util Namespace Reference

Functions
bool	IsCompatibleBroadcastShape (const SizeVector &l_shape, const SizeVector &r_shape)
	Returns true if two shapes are compatible for broadcasting. More...
SizeVector	BroadcastedShape (const SizeVector &l_shape, const SizeVector &r_shape)
	Returns the broadcasted shape of two shapes. More...
bool	CanBeBrocastedToShape (const SizeVector &src_shape, const SizeVector &dst_shape)
	Returns true if src_shape can be brocasted to dst_shape. More...
SizeVector	ReductionShape (const SizeVector &src_shape, const SizeVector &dims, bool keepdim)
	Returns the shape after reduction. More...
int64_t	WrapDim (int64_t dim, int64_t max_dim, bool inclusive=false)
	Wrap around negative dim. More...
SizeVector	InferShape (SizeVector shape, int64_t num_elements)
SizeVector	Concat (const SizeVector &l_shape, const SizeVector &r_shape)
	Concatenate two shapes. More...
SizeVector	Iota (int64_t n)
	Returns a SizeVector of {0, 1, ..., n - 1}, similar to std::iota. More...
SizeVector	DefaultStrides (const SizeVector &shape)
	Compute default strides for a shape when a tensor is contiguous. More...
std::pair< bool, SizeVector >	Restride (const SizeVector &old_shape, const SizeVector &old_strides, const SizeVector &new_shape)

Function Documentation

BroadcastedShape()

SizeVector open3d::core::shape_util::BroadcastedShape	(	const SizeVector &	l_shape,
		const SizeVector &	r_shape
	)

Returns the broadcasted shape of two shapes.

E.g. BroadcastedShape({3, 1, 2}, {5, 1}) -> {3, 5, 2} BroadcastedShape({3, 1, 2}, {5, 3}) -> Exception

Parameters

l_shape	Shape of the left-hand-side Tensor.
r_shape	Shape of the left-hand-side Tensor.

Returns

The broadcasted shape.

CanBeBrocastedToShape()

bool open3d::core::shape_util::CanBeBrocastedToShape	(	const SizeVector &	src_shape,
		const SizeVector &	dst_shape
	)

Returns true if src_shape can be brocasted to dst_shape.

E.g. CanBeBrocastedToShape({1, 2}, {3, 5, 2}) -> true CanBeBrocastedToShape({1, 2}, {3, 5, 3}) -> false

Parameters

src_shape	Source tensor shape.
dst_shape	Destination tensor shape.

Returns

Returns true if src_shape can be brocasted to dst_shape.

Concat()

SizeVector open3d::core::shape_util::Concat	(	const SizeVector &	l_shape,
		const SizeVector &	r_shape
	)

Concatenate two shapes.

DefaultStrides()

SizeVector open3d::core::shape_util::DefaultStrides

(

const SizeVector &

shape

)

Compute default strides for a shape when a tensor is contiguous.

InferShape()

SizeVector open3d::core::shape_util::InferShape	(	SizeVector	shape,
		int64_t	num_elements
	)

Infers the size of a dim with size -1, if it exists. Also checks that new shape is compatible with the number of elements.

E.g. Shape({2, -1, 4}) with num_elemnts 24, will be inferred as {2, 3, 4}.

Ref: PyTorch's aten/src/ATen/InferSize.h

Iota()

SizeVector open3d::core::shape_util::Iota

(

int64_t

n

)

Returns a SizeVector of {0, 1, ..., n - 1}, similar to std::iota.

IsCompatibleBroadcastShape()

bool open3d::core::shape_util::IsCompatibleBroadcastShape	(	const SizeVector &	l_shape,
		const SizeVector &	r_shape
	)

Returns true if two shapes are compatible for broadcasting.

E.g. IsCompatibleBroadcastShape({3, 1, 2}, {5, 1}) -> true IsCompatibleBroadcastShape({3, 1, 2}, {5, 3}) -> false

Parameters

l_shape	Shape of the left-hand-side Tensor.
r_shape	Shape of the left-hand-side Tensor.

Returns

Returns true if l_shape and r_shape are compatible for broadcasting.

ReductionShape()

SizeVector open3d::core::shape_util::ReductionShape	(	const SizeVector &	src_shape,
		const SizeVector &	dims,
		bool	keepdim
	)

Returns the shape after reduction.

E.g. CanBeBrocastedToShape({1, 2}, {3, 5, 2}) -> true CanBeBrocastedToShape({1, 2}, {3, 5, 3}) -> false

Parameters

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

Restride()

std::pair< bool, SizeVector > open3d::core::shape_util::Restride	(	const SizeVector &	old_shape,
		const SizeVector &	old_strides,
		const SizeVector &	new_shape
	)

1. Separate oldshape into chunks of dimensions, where the dimensions are `‘contiguous’' in each chunk, i.e., oldstride[i] = oldshape[i+1] * oldstride[i+1]
2. newshape must be able to be separated into same number of chunks as oldshape was separated into, where each chunk of newshape has matching `‘numel’', i.e., number of subspaces, as the corresponding chunk of oldshape. Ref: aten/src/ATen/TensorUtils.cpp

WrapDim()

int64_t open3d::core::shape_util::WrapDim	(	int64_t	dim,
		int64_t	max_dim,
		bool	inclusive = false
	)

Wrap around negative dim.

E.g. If max_dim == 5, dim -1 will be converted to 4.

Parameters

dim	Dimension index
max_dim	Maximum dimension index
inclusive	Set to true to allow dim == max_dim. E.g. for slice T[start:end], we allow end == max_dim.