RegisterDriver.jl

RegisterDriver.jl drives image registration workflows in the BlockRegistration ecosystem. It runs AbstractWorker algorithms frame-by-frame across an image stack, optionally in parallel across multiple threads, and saves results to disk in JLD format.

Installation

RegisterDriver.jl is distributed through the HolyLab registry. Add that registry before installing:

using Pkg
pkg"registry add https://github.com/HolyLab/HolyLabRegistry.git"
Pkg.add("RegisterDriver")

Concepts

Workers and monitors

A worker is an AbstractWorker instance (from a RegisterWorker* package such as RegisterWorkerApertures) that encapsulates a registration algorithm for a particular compute device. Before running, create a monitor dict that names which computed quantities to collect from each frame:

algorithm = MyWorker(fixed, params...)         # construct an AbstractWorker
mon = monitor(algorithm, (:tform, :mismatch))  # fields to record

The driver

driver iterates the worker over every frame of an image stack. It handles initialisation, per-frame execution, and teardown, then either saves the collected values to a JLD file or returns them in-memory for single-image use.

Usage

Single image (in-memory)

result = driver(algorithm, img, mon)
tform  = result[:tform]

Image stack saved to a file

driver("results.jld", algorithm, img, mon)

Results for each frame are stored inside the JLD file: scalars as plain vectors, bit-type arrays as higher-dimensional HDF5 datasets, and other values inside per-frame "stack<n>" groups.

Parallel multi-threaded registration

Start Julia with multiple threads (e.g. julia --threads=4), then assign one worker per worker thread:

tids       = threadids()
algorithms = [MyWorker(fixed, params...; workertid=t) for t in tids]
monitors   = [monitor(algorithms[1], (:tform, :mismatch)) for _ in tids]
driver("results.jld", algorithms, monitors)

threadids() returns the sorted list of thread IDs that Julia actually schedules @threads tasks on (typically excluding thread 1, which drives the writer).

Loading a device-specific backend

Some workers require a device-specific mismatch package (e.g. a CUDA backend) to be loaded on the driver process before registration starts:

mm_package_loader(algorithm)
driver("results.jld", algorithm, img, mon)

For a full introduction see the RegisterWorkerApertures documentation.