Tour a finished workflow

This tour uses a complete, tested graph. You will inspect it—not rebuild it—so you can learn the interface before dealing with graph editing.

Open the example

Choose:

Open example… → Segmentation & Labels → Red-Channel Label Cleanup

The Image Source is set to VIPP synthetic multichannel volume. Its channels are blue, green, and red; this workflow uses the red/TRITC-like channel.

Full interface context for the tour: the graph runs left to right below the napari viewer, and the selected volume filter is editable in the inspector.

Read the graph from source to result

flowchart LR
  A["Image Source"] --> B["Split Channels"]
  B --> C["Gaussian Blur"]
  C --> D["Otsu Threshold"]
  D --> E["Fill Holes"]
  E --> F["Label Connected Components"]
  F --> G["Clear Border Objects"]
  G --> H["Filter Labels By Volume"]

Each connection carries data from an output port to a compatible input port. The visible graph records processing order, but it does not contain cached pixel arrays.

Inspect one decision at a time

Select each node from left to right:

1. Split Channels: confirm which output is the red signal.
2. Gaussian Blur: compare the source and smoothed thumbnails.
3. Otsu Threshold: inspect the binary foreground mask.
4. Fill Holes: look for changed interiors, not only object outlines.
5. Label Connected Components: confirm touching structures did not become implausible single objects.
6. Clear Border Objects: confirm that edge-touching biology was intentionally excluded.
7. Filter Labels By Volume: inspect the incoming size distribution and the labels that were removed.

Use Pin selected to keep an important output as a napari layer while you select another node. A pinned layer is for comparison; it is not a new graph connection.

Change a parameter safely

Select Gaussian Blur, note its current value, and make a small change. Watch the threshold mask and final labels update downstream. Undo the edit with the toolbar undo action.

This demonstrates a central property of an interactive graph: one parameter can change every downstream result. The ability to see that change helps with review, but it does not tell you which value is biologically correct.

Recognize manual nodes

Measurement, graph-analysis, RACC, and deconvolution nodes can use a manual/cached calculation policy. When one is stale or has not run, select it and choose Calculate or use Calculate all. After upstream changes the button may read Recalculate.

Save a copy

Choose Save workflow… and save the JSON under a new name. Reopen it with Load workflow…. The graph and settings should return; cached images and tables are deliberately recomputed from a source.

Next, build the small graph yourself.