nornir-imageregistration
Core image registration algorithms for aligning 2D images into larger mosaics and 3D volumes.
The alignment walkthrough is Overview of volume building. CPU versus GPU, and where phase correlation, grid refine, and the blob filter live, is CPU and GPU.
Python import:
nornir_imageregistration
Optional GPU extras (see package README.md) are not required for the CPU code paths.
STOS image paths
StosFile.Save stores control/mapped image and mask paths relative to the
``.stos`` directory when a relative form is expressible. StosFile.Load
resolves those lines back to absolute paths in memory. When the image and
.stos locations do not share a common root (cross-drive on Windows, or
incompatible UNC shares), Save falls back to an absolute path and logs a
warning — there is no portable relative encoding across those boundaries.
Use Pyre path replacement / volume reroot when moving trees between machines.
CPU / GPU dual-class transforms
Host and device control-point classes must keep mutate / dedupe / Flip /
FlipWarped semantics aligned. See package design note
nornir-imageregistration/docs/cpu_gpu_dual_class_parity.md and tests in
tests/transforms/test_cpu_gpu_controlpoint_parity.py.
Flip / Flop / mosaic coordinates
MosaicFile.Write: Flip negates Y, Flop negates X. Utah
idoc import writes with Flip=not Flip relative to FlipList.txt.
Full contract: nornir-imageregistration/docs/flip_flop_mosaic_coordinates.md.
API reference