Skip to content

The napari plugin

The napari plugin drives the same registration pipeline interactively: load a dataset, check how the sources are laid out, preview a pair before committing to a full run, and watch each step's progress in napari's own progress bar.

Start napari and open the plugin from Plugins → muvis-align.

The project file

The plugin keeps its settings in a project file (muvis_align_project.yml by default), chosen on the project tab. This is not the general/operations file the command-line pipeline takes - it is one section per tab:

input_output:
  input_path: data/*/*.tiff
  output_path: output
  source_scale_x: '0.004'
  source_position_x: fn[-2]*24
  overwrite: true
pre_processing:
  scale: 8
  normalisation: none
registration:
  operation: register
  method: phase_correlation
  transform_type: rigid
fusion:
  method: average
  spacing: mean
  ome_version: '0.5'

Every widget writes straight back to this file as you change it, so a project can be reopened exactly as it was left. Paths are stored relative to the directory the project file sits in.

Using the plugin

Step by step in napari goes through the tabs in order, with every parameter and what each button does.

Progress

Each operation reports one progress bar, following the work itself rather than the number of dask tasks, so the bar does not restart partway through. Where a step's progress cannot be observed directly - the global optimisation in particular, which previously sat at 0% for the whole run - it is followed through its own log output and counted against the number of edges it can remove.

Exporting reports two bars in sequence: one for the export, then one for drawing the result.

Performance notes

  • A project opens without fusing anything: the layout view is drawn from the source geometry, read from metadata alone.
  • Leave Block size at Automatic unless the on-disk layout matters. Fusion's blocks share a quarter of the memory actually available to the process - on a cluster the job's allocation, not the node's - each waiting for room rather than all being sized for the busiest one.
  • Sources without a usable pyramid make every preview expensive. Convert them first.
  • The Target downscale factor on the pre-processing tab is usually the largest single win when tuning registration: it applies to registration only, so the exported result stays at full resolution.