This page lists all 146 operation specifications registered by
NODE_LIBRARY in 0.16.0a7. The palette exposes 144: the two
legacy scatter raster operations remain loadable but are hidden from new-node
selection. The 0.16.0a1 additions are Grow Regions from Seeds — CellProfiler
Propagation and six CellProfiler compartment stages;
0.15.0a5 has 126 specifications and 124 palette nodes.
New in 0.16.0a3
Template Match, Find Peaks, and Detect Spots per Frame are
CPU nodes under Image Data → Detection; Build Tracks is under
Image Data → Tracking. All four are included in the release counts below.
New in 0.16.0a4
Review Images is a presentation-only two-input sink with Open review…,
a linked side-by-side/overlay window and saved display recipes. It has no
scientific image output and is included in the counts above. See
the review reference. Compare Images remains the
separate quantitative comparison node. ImageJ Gaussian Blur adds the
bounded CPU ImageJ 1.54p compatibility operation under Smoothing & Denoising.
New in 0.16.0a2
Five additions are included in the counts above: Estimate
Registration, Apply Transform, Compare Images, Skeletonize
Labels and Analyze Skeleton per Label. The registration nodes are
described below.
Skeletonize Labels and Analyze Skeleton per Label are marked
separately below. They preserve original object IDs for morphology joins;
see the skeleton reference.
Existing Label Skeleton Components behavior is unchanged.
Save Image accepts images or 3D meshes and passes the connected
image/mesh type downstream. Its format menu and Batch Output's menu follow
the connected input; see mesh saving.
Scope of this reference
Titles and families follow the 0.16.0a7 source registry.
Input and output summaries describe the ordinary/default ports; Split Channels,
Split Axis, Born-Wolf PSF, and other multi-output nodes can resolve ports
from runtime data. Parameter widgets, defaults, and bounds in the installed
release remain the source of truth until a generated parameter reference is
published.
Since 0.14.0a1, manual/cached nodes and nodes selected for isolated tuning use
the same graph-wide execution language.
Bright amber identifies the node that needs action; dark amber identifies
downstream nodes that are stale but waiting for that action. Tune node in
isolation recalculates only the selected node until
the session is applied or cancelled. This changes execution and presentation,
not the operation's scientific definition.
Implemented nodes can also show an actual-run CPU, GPU · CuPy, or amber
CPU fallback badge. The badge is execution provenance for
the accepted result, not part of the operation's scientific definition. A GPU
choice appears for an operation with a declared provider; it is not filtered by
the current dtype, parameters, shape, memory, dependencies, or environment.
Call-specific admission happens during planning and can select CPU, visibly
fall back, or fail. See the
0.15.0a3 CPU/GPU operation matrix
for the accelerated node families and their current public regions.
Build Tracks consumes a complete calibrated observation series from Detect
Spots per Frame or time-series object measurements. It outputs Tracked
observations and Track summary tables with one-to-one links, explicit gaps and review
flags. See Track objects and spots and
tracking rules. It does not infer divisions or fusions.
Crop explicit Y/X and optional Z edges while preserving T/C, graph-port type, calibration, and origin; an eligible sole direct local OME-Zarr source can read only the retained window.
Select Axis Slice
array
any
Keep or select an index/range along an axis.
Split Axis
array
dynamic outputs
Split time, z, or other non-channel axes.
Reorder Axes
array
same as input
Transpose pixels and their existing axis records into a reviewed order; does not rename axes.
Set Pixel Size / Units
array
same as input
Repair or define physical calibration.
Set Microscope Metadata
array
same as input
Record missing channel emission wavelengths, objective numerical aperture, or immersion refractive index without changing pixels.
Composite → RGB separates Channel axis mode and RGB mapping mode.
Each can be Auto, with its resolved choice shown read-only, or Manual. Manual
axis mode permits any valid deliberate axis choice. Manual mapping provides a
dynamic selector for every source channel: Unassigned, Red, Green, Blue,
Magenta, Cyan, or Yellow. Unassigned channels contribute nothing; all assigned
channels, including stacks wider than RGB/RGBA, mix additively. Changing the
mapping invalidates the composite and its descendants only. Every calculated
manual result upstream remains cached in every cache mode; unchanged automatic
intermediates remain eligible for normal Smart/Low-memory pruning. Intensity
mapping is a separate native-preserving versus explicitly lossy percentile
choice.
Clamp values to the chosen lower and upper bounds.
Rescale Intensity offers exact full-input percentiles or saved explicit input
values. Clamp Intensity offers an unchanged full-data-range mode or saved explicit clamp
values. Integer percentiles and clipping preserve native int64/uint64 levels;
unrepresentable fractional bounds or intervals too wide for faithful ratio
arithmetic report an error rather than being silently rounded. Auto Contrast
for Linear Scale + Offset writes calculated parameters
and is not merely a display adjustment. See
Tune without fooling yourself.
These nodes have been public since 0.14.0a1 and remain available in 0.15.0a2.
They have synthetic examples and automated coverage, but broad real-image
restoration validation remains an evidence gap;
see validation status.
Node
Input
Output
Execution
Use
Born-Wolf PSF
image/reference metadata
image
automatic
Generate a PSF from metadata or parameter overrides.
Otsu, Triangle, Yen, Isodata, and Minimum count every finite input value.
Integers use exact native levels for observed spans up to 65,536; floats use the
saved 2–65,536 bin setting (default 256). Li operates on raw finite values.
Minimum also saves its maximum smoothing iterations and fails explicitly when
it cannot identify a valid two-peak histogram. See
Segmentation and label cleanup.
All eight nodes are CPU operations, manual/cached by default. Mesh outputs are
inspected as napari surfaces, not graph thumbnails. See the
mesh-object tutorial for grouping, calibration,
refinement caveats and 3MF/OBJ export.
Group
Node
Input → output
Use
Create surfaces
Mask to 3D Mesh
Boolean 3D mask → mesh
One foreground object or separate face-connected objects.
Create surfaces
Labels to 3D Mesh
3D labels → mesh
One mesh object per positive label, retaining IDs.
Objects & colours
Colour Mesh Objects
mesh → mesh
Categorical IDs or current-geometry measurement colours.
Objects & colours
Combine Meshes
2–16 meshes → mesh
Collect objects and colours in one calibrated frame; no union or registration.
Objects & colours
Split Mesh Objects
mesh → mesh
Separate shared-edge components; cavity walls split too.
Objects & colours
Filter Mesh Objects
mesh → mesh
Keep objects inside/outside a measurement or ID range.
Finish Cell Regions — CellProfiler (new in 0.16.0a1)
grown regions plus retained nuclei
labels
Fill labelled holes and map cells to retained nucleus IDs.
Extract Cytoplasm — CellProfiler (new in 0.16.0a1)
cells plus nuclei
labels
Subtract nuclei, optionally retaining their one-pixel outline.
Label Connected Components
mask
labels
Convert mask to object IDs.
Clear Border Objects
mask or labels
same type
Remove objects touching image boundary.
Filter Labels By Volume
labels
labels
Keep/remove labels by pixel/voxel count.
Filter Labels By Property
labels plus table
labels
Filter labels using table columns.
Relabel Sequential
labels
labels
Compact labels after filtering.
Find Label Boundaries (new in 0.15.0a3)
labels or Boolean mask
Boolean mask
Same-grid boundary QC with inside/outside/both placement; no retained IDs or mesh.
Label Skeleton Components
mask
labels
Label connected skeleton components.
Label Skeleton Branches
mask
labels
Label skeleton branch paths.
See Inspect label boundaries
for placement, connectivity, spatial scope and crop-border caveats. Find Label
Boundaries runs on CPU and preserves the input calibration.
Load collected results with their retained types, units and identity records.
Merge Tables
tables
table
Join measurement branches.
Add Metadata Columns
table
table
Add treatment, replicate, condition, batch, or other fields.
Select Table Columns
table
table
Keep and reorder columns.
Summarize Measurements / Statistics (name and recipe new in 0.16.0a1)
table
table
Descriptive summaries; new recipes explicitly distinguish objects, image means and sample means with weighting, exclusions and counts.
Plot Results (new in 0.16.0a1)
table
plot
Compare individual/grouped points, histograms, cumulative distributions and scatter plots; open an editable window and export a sized figure.
See Plot measurement results for
single-image examples, image means, exclusions and export. Plot Results is
descriptive; it does not add statistical tests. See
Summarize measurements for Statistics;
older summaries retain their legacy calculations until explicit upgrade.