torchrelay.extra.bounds

Boundary conditions

There is no common convention to name boundary conditions. This file lists all possible aliases and provides tool to "convert" between them. It also defines function that can be used to implement these boundary conditions.

NITorch	SciPy	PyTorch	Other	Description
replicate	border	nearest, replicate	repeat	a a | a b c d | d d
zero	constant(0)	constant, zero	zeros	0 0 | a b c d | 0 0
dct2	reflect	reflection (align_corners=True)	neumann	b a | a b c d | d c
dct1	mirror	reflection (align_corners=False)		c b | a b c d | c b
dft	wrap	circular		c d | a b c d | a b
dst2			antireflect, dirichlet	-b -a | a b c d | -d -c
dst1			antimirror	-a 0 | a b c d | 0 -d

to_nitorch

to_nitorch(bound)

Convert boundary type to NITorch's convention.

Parameters:

Name	Type	Description	Default
bound	[list of] str or bound_like	Boundary condition in any convention	required
as_enum	bool	Return BoundType rather than str	False

Returns:

Name	Type	Description
bound	[list of] str or BoundType	Boundary condition in NITorch's convention

to_scipy

to_scipy(bound)

Convert boundary type to SciPy's convention.

Parameters:

Name	Type	Description	Default
bound	[list of] str or bound_like	Boundary condition in any convention	required

Returns:

Name	Type	Description
bound	[list of] str	Boundary condition in SciPy's convention

to_torch

to_torch(bound)

Convert boundary type to PyTorch's convention.

Parameters:

Name	Type	Description	Default
bound	[list of] str or bound_like	Boundary condition in any convention	required

Returns:

Type	Description
[list of]	bound : str Boundary condition in PyTorchs's convention align_corners : bool or None

replicate

replicate(i, n)

Apply replicate (nearest/border) boundary conditions to an index

Aliases

border, nearest, repeat

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	{1, 0, -1}	Sign of the transformation (always 1 for replicate)

dft

dft(i, n)

Apply DFT (circulant/wrap) boundary conditions to an index

Aliases

wrap, circular

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	{1, 0, -1}	Sign of the transformation (always 1 for dft)

dct2

dct2(i, n)

Apply DCT-II (reflect) boundary conditions to an index

Aliases

reflect, neumann

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	{1, 0, -1}	Sign of the transformation (always 1 for dct2)

dct1

dct1(i, n)

Apply DCT-I (mirror) boundary conditions to an index

Aliases

mirror

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	{1, 0, -1}	Sign of the transformation (always 1 for dct1)

dst1

dst1(i, n)

Apply DST-I (antimirror) boundary conditions to an index

Aliases

antimirror

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	[tensor of] {1, 0, -1}	Sign of the transformation

dst2

dst2(i, n)

Apply DST-II (antireflect) boundary conditions to an index

Aliases

antireflect, dirichlet

Parameters:

Name	Type	Description	Default
i	int or tensor	Index	required
n	int	Length of the field of view	required

Returns:

Name	Type	Description
i	int or tensor	Index that falls inside the field of view [0, n-1]
s	[tensor of] {1, 0, -1}	Sign of the transformation (always 1 for dct1)