API Summary

Geometries

kernelkit.VolumeGeometry(shape, voxel_size, ...)	Geometry for a single 3D reconstruction object.
kernelkit.ProjectionGeometry(...[, u, v, ...])	Single 3D projection with one source and one detector.
kernelkit.Beam(value[, names, module, ...])	Beam type.
kernelkit.resolve_volume_geometry([shape, ...])	Resolves a VolumeGeometry given partial parameters.
kernelkit.experimental.suggest_volume_extent(...)	Suggest a volume extent for a given geometry.
kernelkit.Detector(rows, cols, pixel_height, ...)	Detector specifications.
kernelkit.rotate(geom[, roll, pitch, yaw])	Creates a new geometry by rotation.
kernelkit.rotate_(geom[, roll, pitch, yaw])	In-place rotation of a geometry.
kernelkit.scale(geom, factor)	Creates a new projection geometry by factor.
kernelkit.scale_(geom, factor)	In-place factor of geometry or geometry sequence.
kernelkit.shift(geom, vector)	Creates a new geometry by a 3D shift.
kernelkit.shift_(geom, vector)	In-place shift of a geometry or geometry sequence.

Kernels

kernelkit.kernel.BaseKernel(cuda_source)	Abstract base class for CUDA kernels
kernelkit.kernel.copy_to_texture(array[, ...])	Creates a single-channel 2D/3D texture object of type float from a CuPy array.
kernelkit.kernel.copy_to_symbol(module, ...)	Copy array to address on GPU, e.g. constant memory.
kernelkit.kernels.VoxelDrivenConeBP([...])	Voxel-driven conebeam backprojection kernel.
kernelkit.kernels.RayDrivenConeFP([...])	Ray-driven conebeam forward projection kernel.

Projectors

kernelkit.BaseProjector()	Interface-like base class for projectors.
kernelkit.BackProjector([kernel, texture_type])	Backprojector for conebeam geometry
kernelkit.ForwardProjector([kernel, ...])	Cone-beam forward projector.
kernelkit.toolbox_support.ForwardProjector()	Adapter for ASTRA forward projectors.
kernelkit.toolbox_support.BackProjector()	Adapter for ASTRA backprojectors.

Algorithms

Reference algorithms:

kernelkit.bp(projections, ...[, filter, ...])	X-ray backprojection.
kernelkit.fp(volume, projection_geometry, ...)	X-ray forward projection.
kernelkit.fdk(projections, ...[, filter])	Feldkamp-Davis-Kress algorithm for conebeam geometry.
kernelkit.sirt(projections, ...[, iters, ...])	Simultaneous Iterative Reconstruction Technique.

Operator-style algorithm building blocks:

kernelkit.BaseOperator()	A mathematical operator.
kernelkit.ProjectorOperator(projector, ...)	An operator \(A : X \to Y\) using a forward and backprojector.
kernelkit.XrayTransform(projection_geometry, ...)	A mathematical operator representing the X-ray transform.

Miscellaneous

Helpers for cone and Fourier-based algorithms:

kernelkit.processing.filter(projections[, ...])	Filter in Fourier domain
kernelkit.processing.preweight(projections, ...)	Pixelwise rescaling to compensate for ray length in conebeam images

Helpers for pitched 2D memory:

kernelkit.data.ispitched(array)	Returns True if the array is pitched, False otherwise.
kernelkit.data.aspitched(array[, xp])	Pads array to pitched shape and returns a view in original shape.

Helpers for PyTorch:

kernelkit.torch_support.AutogradOperator(...)

Autograd function for Operator objects.