Alignment configuration
Configurations, plans and checkpoints behind tj.align.
tomojax.alignment provides the building blocks behind tj.align and
tomojax align.
AlignConfig: every solver setting. Its fields are also the keys of atomojax align --configTOML file.alignment_plan(mode, grid, quality=..., levels=..., freeze=...): the configuration and coarse-to-fine levelstj.alignruns for a mode. Change a field withdataclasses.replace(plan.config, ...)and pass the result astj.align(scan, mode=..., config=...).align_multires: the array-level solver.
tj.align(scan, checkpoint="align.ckpt") saves progress after each outer
iteration and resumes an interrupted run from it. Without expert overrides,
tj.align chooses centre estimation for axial parallel scans and guarded
setup, support and motion selection for unposed laminography; its result records
conditional outcomes and limitations. Explicit mode="pose" selects per-view
corrections (five parameters, six in a cone beam). Smooth pose models, setup
stages and mixed schedules remain available explicitly. See the
alignment guide.
Joint alignment retries without its derivative cache before skipping a level
that does not fit GPU memory. The result records factors_skipped and
factors_skipped_reason; a completed checkpoint retains that reason on resume.
Import from tomojax.alignment or tomojax.alignment.api.