Channel and axis controls¶
Start with axes and metadata to establish what each dimension means. These controls make deliberate changes; they do not infer biological channel identity from dimension length.
Choose the operation¶
| Need | Node |
|---|---|
| One scalar channel | Extract Channel |
| Separate graph outputs for every semantic channel | Split Channels |
| Separate time points, Z slices, or another non-channel axis | Split Axis |
| A channel-last RGB image | Composite → RGB |
| Resize spatial dimensions | Rescale Axes |
| Supply missing acquisition facts | Set Microscope Metadata |
Composite → RGB¶
Axis choice and colour assignment are independent:
| Control | Auto | Manual |
|---|---|---|
| Channel axis mode | Uses the explicit carried channel axis; the resolved choice is read-only. | Lets you choose a valid axis deliberately, including Z; the choice is recorded. |
| RGB mapping mode | Shows the resolved mapping from encoded RGB/RGBA or fluorescence pseudo-colours. | Exposes one saved colour choice per detected source channel. |
Missing or ambiguous channel semantics are errors in Auto, not an invitation to guess from a trailing dimension of length three or four.
Manual colours are Unassigned, Red, Green, Blue, Magenta, Cyan, and Yellow. Composite colours contribute to multiple RGB planes; multiple channels can add to the same plane. Unassigned contributes nothing. There is no three- or four-source-channel limit.
Auto mapping:
- preserves declared RGB/RGBA order and ignores alpha;
- blends all fluorescence channels using their carried pseudo-colours;
- falls back to Blue, Green, Red, Magenta, Yellow, Cyan, repeating as needed;
- copies one scalar channel into all three RGB planes.
Intensity mapping is a separate scientific choice:
- Preserve numeric values keeps native scale without normalization or clipping and rejects unsafe precision/overflow cases.
- Per-channel 1st–99th percentile (lossy) normalizes channels independently
and clips additive mixtures to
[0, 1].
Changing the mapping invalidates this node and its descendants, not unchanged upstream manual results. Smart/Low-memory cache modes may still prune automatic intermediates under their normal retention policy.
Which Split Channels output is inspected?¶
When exactly one distinct output is connected, that channel drives the split node's thumbnail, inspect/pinned layer, histogram, metadata, dimensions, and Save selected output…. Several consumers of the same port still count as one.
With no connected output, or several different outputs in use, these surfaces use the saved Thumbnail channel. This presentation choice neither rewires the graph nor changes the arrays delivered downstream.
Rescale Axes with inferred names¶
Unique inferred X/Y axes may be resized after reviewing the inspector's amber warning. A real X/Y size change records that plane as explicit; a no-op does not change confidence. Other inferred axes retain their metadata.
This does not infer depth. A QYX image keeps Q unchanged until you explicitly
establish that Q means Z. An inferred Z axis must likewise be declared before
a nontrivial Z resize. Interactive, exported, and batch execution use the same
rule.
Set Microscope Metadata¶
This pass-through node supplies up to three channel emission wavelengths, objective numerical aperture, and immersion refractive index. Zero means leave the carried value unchanged, not erase it.
Channels 2/3 require an explicit channel axis and a matching channel count. Values travel with the image and its history without changing pixels. This is separate from Set Pixel Size / Units, which supplies spatial calibration.