Data Types

VIPP uses typed graph data to communicate and check intended scientific meaning. Port types prevent many category errors; they cannot determine whether an image, mask, or table is appropriate for your biological question.

Type	Meaning	Typical napari display
image	Intensity image, RGB image, PSF, or restored image.	Image layer
mask	Binary foreground/background image.	Labels-style overlay
labels	Integer object IDs, with zero as background.	Labels layer
table	Measurement or summary rows.	Inspector table preview and CSV/TSV output
array	Generic array accepted by many processing nodes.	Depends on output node
mask_or_labels	Input may be either a binary mask or label image.	Depends on connected input
any	Pass-through or generic output where type is resolved from context.	Depends on connected input

Images

Images carry array data plus metadata such as axes, scale, unit, channel names, source identity, and operation history where available.

Convert Dtype is an explicit scientific operation. Its scaling choice can preserve representable values or intentionally remap the range, which may change thresholds, rounding, writers, memory, and downstream interpretation. VIPP's optimizer never inserts a cast to make a GPU candidate eligible. Where float32 is justified, an authored conversion can unlock much larger Gaussian or deconvolution gains; record and validate that conversion like any other node.

Masks

Masks represent inside/outside decisions. They are usually produced by thresholding, morphology, or colocalized-voxel nodes.

Use masks before label creation when you are still deciding what is foreground.

VIPP stores binary masks as Boolean arrays where possible. If a Boolean mask is connected to a global automatic-threshold node, VIPP recognizes that it is already segmented and preserves its True/False decisions. It does not reinterpret the two levels through Otsu, Triangle, Li, Yen, Isodata, or Minimum thresholding.

For floating-point intensity images, non-finite pixels (NaN, positive infinity, or negative infinity) are excluded from global automatic-threshold estimation and become False in the resulting mask. Inspect why those values exist; treating them as background is a defined calculation rule, not evidence that the source data are valid.

Labels

Labels represent object identity. Label 0 is background; positive integers are objects.

Use labels when you want object measurements, label cleanup, object colocalization, or event assignment.

Tables

Tables are first-class outputs, not fake images. They appear in the inspector and can be saved as CSV or TSV.

Common table nodes include:

• Measure Objects
• Measure Objects + Intensity
• Measure 3D Mesh Morphology
• skeleton and network measurements
• colocalization metrics
• Merge Tables
• Add Metadata Columns
• Summarize Measurements

Manual Calculation Nodes

Expensive measurement, graph-analysis, RACC, and deconvolution nodes can be manual/cached. Select the node and click Calculate or use toolbar Calculate all.