3D Data Objects

Volume and projection data

ASTRA data objects come in two varieties: volume data and projection data (sinograms). A data object can be created using the following commands:

create

id = astra.data3d.create('-vol', vol_geom)
id = astra.data3d.create('-vol', vol_geom, initializer)

id = astra_mex_data3d('create', '-vol', vol_geom);
id = astra_mex_data3d('create', '-vol', vol_geom, initializer);

Create an initialized 3D volume data object for the geometry vol_geom.

Initializer may be:

a scalar: the object is initialized with this constant value.
a matrix: the object is initialized with the contents of this matrix. The matrix must be of size (n_slices, n_rows, n_cols) in Python, or (n_cols, n_rows, n_slices) in MATLAB, as defined by the volume geometry. In Python, it must be convertible to float32. In MATLAB, it must be of class single, double or logical.

If no initializer is given, the volume is initialized to all zeros.

id = astra.data3d.create('-proj3d', proj_geom)
id = astra.data3d.create('-proj3d', proj_geom, initializer)

id = astra_mex_data3d('create', '-proj3d', proj_geom);
id = astra_mex_data3d('create', '-proj3d', proj_geom, initializer);

Create an initialized 3D projection data object for the geometry proj_geom.

Initializer may be:

a scalar: the object is initialized with this constant value.
a matrix: the object is initialized with the contents of this matrix. The matrix must be of size (det_row_count, n_angles, det_col_count) in Python, or (det_col_count, n_angles, det_row_count) in MATLAB, as defined by the projection geometry. In Python, it must be convertible to float32. In MATLAB, it must be of class single, double or logical.

If no initializer is given, the projection data object is initialized to all zeros.

link

id = astra.data3d.link('-sino', proj_geom, array)
id = astra.data3d.link('-vol', vol_geom, array)

Create an ASTRA data object that shares its memory with the specified CPU or GPU array supporting the DLPack interface. This includes arrays from NumPy, PyTorch, JAX, TensorFlow and CuPy libraries. The array must be contiguous, have float32 dtype, and a shape compatible with the specified geometry. No copying is performed, so changes to the linked array will be visible to ASTRA, and vice versa. Linking the array increments the reference count of the underlying memory, so deleting the original array will not destroy the created ASTRA object.

Added in version 2.2: Accept any 3D array supporting DLPack protocol. Before, only NumPy arrays were supported.

Phantom generation

ASTRA includes a built-in generator for the 3D Shepp-Logan phantom:

shepp_logan

Added in version 2.2.

id, data = astra.data3d.shepp_logan(vol_geom, modified)

Create a Shepp-Logan phantom as an ASTRA volume data object (id), and a NumPy array (data). modified=True creates a phantom with improved contrast (default).

API

get

A = astra.data3d.get(id)

Fetch the data object as a 3D array with dtype float32.

A = astra_mex_data3d('get', id);

Fetch the data object as a 3D matrix of class double.

get_shared

A = astra.data3d.get_shared(id)

Fetch the data object as a 3D NumPy array sharing its memory with the ASTRA object. Changes to the returned array will be visible to ASTRA, and vice versa. Deleting the ASTRA object while the resulting Python object still exists will lead to undefined behaviour and potentially memory corruption and crashes.

get_single

Only available in MATLAB interface.

A = astra_mex_data3d('get_single', id);

Fetch the data object as a 3D matrix of class single.

set / store

astra.data3d.store(id, A)

astra_mex_data3d('set', id, A);
astra_mex_data3d('store', id, A);

This stores the matrix A in the data object. The dimensions of A must match the dimensions of the existing data object.

Set and store are synonyms in the MATLAB interface.

dimensions

s = astra.data3d.dimensions(id)

s = astra_mex_data3d('dimensions', id);

Get the dimensions of a data object.

get_geometry

geom = astra.data3d.get_geometry(id)

geom = astra_mex_data3d('get_geometry', id);

Get the (volume or projection) geometry attached to the object.

Note

The returned geometry may slightly deviate from the original geometry because of internal data type conversions.

change_geometry

astra.data3d.change_geometry(id, geom)

astra_mex_data3d('change_geometry', id, geom);

Change the (volume or projection) geometry attached to the object. It cannot change the dimensions of the data object, but can be used to change voxel dimensions or projection angles, for example.

delete

astra.data3d.delete(id)
astra.data3d.delete([id1, id2, ...])

astra_mex_data3d('delete', id);

Free the memory of a data object.

clear

astra.data3d.clear()

astra_mex_data3d('clear');

Free all data objects.

info

astra.data3d.info()

astra_mex_data3d('info')

Print basic information about all allocated data objects.

Linking data in MATLAB

id = astra_mex_data3d('link', '-vol', vol_geom, array, readonly, Z);
id = astra_mex_data3d('link', '-proj3d', proj_geom, array, readonly, Z);

This creates a data object that directly uses a MATLAB array as storage instead of allocating its own memory. The array must be of the same dimensions as those required for initializers in astra_mex_data3d(‘create’). Additionally, it must be of class ‘single’.

The optional argument ‘readonly’ (default: false), controls the exact behaviour of this operation. See the two sections below for details.

The optional argument ‘Z’ (default: 0) allows creating a data object that is smaller in the third dimension than the MATLAB array. The data object will be mapped to slices starting at slice Z. NB: Z is zero-based, unlike MATLAB array indexing.

Read-only link mode:

The data object becomes an additional reference to the array, effectively behaving the same as a MATLAB assignment ‘internal_data = A;’ (if ‘A’ is passed as the ‘array’ argument). If the array A is changed inside MATLAB, a copy will be made and the changes to A will not be visible to this data object.

The data object’s read-only state is not enforced by the astra toolbox. Using it as output for algorithms is allowed, but the exact effects depend on MATLAB’s internal reference counting mechanics.

Read-write link mode:

Warning

Functionality required for this link mode was removed from MATLAB since version 2020a.

The passed array is ‘unshared’ and the data object obtains a second reference to this array. There is no direct MATLAB script equivalent to this, but effectively the data object and the passed array will share memory. Any changes to the data object from inside the toolbox will be visible in MATLAB.

If the passed array is modified in MATLAB, this link is broken (by MATLAB’s reference counting mechanism), and the changes will not be visible to the astra data object.