ct_scan op• Data kinds: projstack × ctgeometry → voxel
• Call: import fullseye as fs; fs.ledger.os_sart_reconstruct(projections, geometry, n_iter=10, n_subsets=16, relaxation=0.8, initial=None, nonnegative=True, device='auto', on_iter=None) (to call the implementation directly, import ctscan; ctscan.os_sart_reconstruct(projections, geometry, n_iter=10, n_subsets=16, relaxation=0.8, initial=None, nonnegative=True, device='auto', on_iter=None); from the registry, opstomography.get("os_sart_reconstruct"))
• GPU: none (CPU reference implementation only)
OS-SART: the SIRT update applied to one ordered subset of views at a time.
One iteration passes over all subsets, so it costs about one SIRT iteration and moves about *n_subsets* times as far early on. With inconsistent data it settles into a limit cycle rather than a point.
:param projections: (view, V, U) line integrals. :param geometry: dict from ct_scan_geometry. :param n_iter: passes over all subsets, 1 .. 5000. :param n_subsets: subsets, 1 .. n_views. :param relaxation: step lam, (0, 2). :param initial: starting volume; None gives zeros. :param nonnegative: clip to >= 0 after each subset. :param device: "auto", "cpu" or "cuda". :param on_iter: optional on_iter(iteration, volume), called after each full
pass; returning a boolean True (bool or numpy.bool_) stops
there, any other return value is ignored.
:returns: (Z, Y, X) float volume. :raises ValueError: on a shape that disagrees with the geometry, a relaxation
outside (0, 2), or more subsets than views. Empty input (projections
with 0 elements) raises ValueError.
The default 10 passes is not converged. On consistent data (24^3 Shepp-Logan, cone, 20 views, 16 subsets) the relative residual is 0.0135 at 10 passes and 0.0044 at 40, nRMS to the truth 0.0278 -> 0.0176 (37 %); on the measured walnut (fan, 50 views, 10 subsets) the RMSE to the reference is 0.0048 /mm at 10 passes and 0.0036 at 40 (2026-11-09 audit). Pick the count from a curve (on_iter, see sirt_reconstruct).
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• ct_scan_modes — py -3.11 examples/ct_scan_modes.py
• poc_ct_casting_sparse_views — py -3.11 examples/poc_ct_casting_sparse_views.py
• poc_real_walnut_cbct — py -3.11 examples/poc_real_walnut_cbct.py
voxel as input)radon_volume · ct_forward_project
ct_scan)ct_scan_geometry · shepp_logan_3d · ellipsoid_phantom · ellipsoid_projections · ct_forward_project · ct_adjoint_project · fdk_reconstruct · sirt_reconstruct
*Provenance: ctscan.py — TOMOGRAPHY operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.