Overview
Nornir is a collection of Python packages for building image mosaics and 3D volumes from serial-section imaging data. Components cover thread/process pools, shared utilities, image registration, and volume construction.
Mission
Nornir’s goal is to take large sets of overlapping images in 2D and 3D and produce registered (aligned) 2D and 3D volumes at any practical size and scale.
Theory and figures: for a guided walkthrough of mosaic capture, tile alignment, slice-to-slice registration, and volume mapping—with the original figures from the public manual—see Overview of volume building.
Documentation convention: narrative and API reference for the umbrella project live in this monodoc. Each nornir-* package also has a short README.md at its repository root with an introduction and links back here.
History
Nornir evolved from a collaboration between the Marc Lab and the Scientific Computing Institute (Tasdizen and Whitaker groups) at the University of Utah. The original tools, known as the NCR Toolset, were used to construct RC1, a 250 µm diameter, 33 µm tall cylinder of rabbit retina at a resolution of 2.18 nm/pixel.
Nornir is a work in progress. It supports importing images from transmission electron microscopes running SerialEM (.idoc files), light microscopes running Surveyor (.pmg files), and other volumes represented with a single image per section (.png files).
System notes
What a machine needs to run Pyre, a workstation pipeline, or Docker is in Host requirements. Production volumes (NAS, scratch disk, GPU class) are in Production envelope.
Terminology
Terms used across the manual (block, section, mosaic, tile, channel, filter, prune, blob, STOS, grid, slice-to-volume) are defined in Concepts.