VIPP 0.16.0a4¶
VIPP 0.16.0a4 adds linked Review Images views, bounded ImageJ Gaussian Blur compatibility and focused fixes for Batch/Results controls, independent Windows workspace windows, threshold inspectors and OME readers/writers.
Use the canonical GitHub release for the exact wheel, source archive, Windows installer, separate Apple Silicon/Intel macOS packages and checksums, or the exact PyPI version. Install only those release assets and verify their hashes. Desktop installers are unsigned alpha builds; macOS packages are also unnotarized and CPU-only. Follow the installation instructions.
Review aligned images together¶
Connect current Image A and optional Image B to Review Images, then choose Open review…. Compare raw/processed intensities, red/green channels, an intensity with its mask/labels, or explicit RGB with a numeric colour map. The compact node has no output image or empty preview. The separate window provides framed pane cards and a familiar inspector sidebar.
Choose Side by side or Overlay, select A/B/A + B in each pane and use shared time/slice controls. Link viewpoint synchronizes camera navigation. Axes and scale-bar visibility apply to both panes; Oblique, XY, XZ, YZ and Fit views sit together below the images. Switching 3D to 2D returns to XY/Z navigation; the editable position and total count are separate.
Per-input contrast, colour maps, opacity, visibility, fixed colour scales and scalar 3D rendering are presentation settings, saved with the workflow without invalidating scientific results. Black/white limits can be locked; a visibility cutoff hides displayed values without calculating a threshold mask.
Keep review open while editing upstream: the last complete pair stays visibly noncurrent while work changes. Accepted current A/B results update together, preserving camera/navigation and display styles on the same grid. Failed, cancelled or superseded reviewed inputs cannot become current. Grid, input-kind or connected-input-presence changes require reopening; there is no silent resampling. Normal memory guards still apply, including in Low-memory mode.
The four Image Review examples raise the catalogue to 35 examples and 34 bundled samples. They include scalar channels, anisotropic physical spheres, categorical label time series and RGB with a constructed 0–1 index. The index is not a computed RACC result. See Review images in linked views and the input/rendering reference.
MIP ignores depth; depth-weighted intensity does not provide opaque occlusion; isosurface lighting changes apparent brightness. RGB 3D uses explicitly labelled additive component MIPs, not voxelwise RGB ray-casting or napari-racc shader parity. RGBA remains 2D-only. Review is display-only, not quantitative validation, segmentation or registration.
Reproduce a bounded ImageJ Gaussian step¶
ImageJ Gaussian Blur targets ImageJ 1.54p's direct Gaussian algorithm on scalar uint8, uint16 and finite float32 images. Sigma (pixels) ranges from 0 through 8.5; leading Z/time/channel planes are independent. Larger sigma, encoded RGB, physical-unit sigma, volumetric blur and GPU execution are outside this node's contract. Ordinary Gaussian Blur and Gaussian Blur 3D retain their algorithms.
Thirty-seven independently executed ImageJ reference cases and a bounded acquired-workflow comparison support this compatibility region, not every Fiji macro or biological analysis. See Fiji/ImageJ compatibility.
Clearer controls and independent windows¶
- Batch output checks explain blocked destinations and offer Find problem to select the affected node. Repair the indicated path, format or filename and recheck; selections do not silently change run scope.
- Select all / Deselect all controls share placement, outlined icons and visible scope labels across Batch, Results, columns and Statistics. Batch Overrides deselection clears hidden selections too; other selectors retain their documented scopes.
- On Windows, detached VIPP maximizes/restores once on title-bar double-click and minimizes independently from napari. Batch and Results have native window controls, support passive-toolbar double-click and retain their prior maximized state when reopening a minimized window. All remain in the same session; ending the session closes its windows. Native Linux/macOS ownership and minimization remain distinct qualification work.
- Manual threshold controls use connected intensity units, including floats outside 0–1. Histogram/guide edits synchronize numeric controls and saved cutoffs. Numeric entry can exceed a practical slider window.
- Wrapped GPU guidance remains readable in narrow inspectors; layer-list repaint recovery avoids repeated loading-indicator errors when rows disappear.
- Older OME-TIFF micrometer/micrometre unit names retain numeric calibration and source pixels.
- Local OME-Zarr 0.4/0.5 image and analysis-dataset exports work with
ome-zarr0.20 and newer, retaining channel display metadata, calibration and VIPP history. The writer compatibility fix does not claim complete metadata fidelity; reopen important outputs in the receiving tool.
See Batch processing, Results Workspace, interface controls and intensity thresholds. The manual home also acknowledges NRF Thuthuka support.
Compatibility and qualification boundaries¶
The catalogue has 146 operation specifications and 144 palette nodes; two legacy scatter raster operations remain loadable but hidden. Workflow, batch-config and manifest schemas remain 6, but new operations still require a compatible runtime. Preserve original workflows, inputs, environments and decisive results. Recalculate representative data and regenerate Python exports with the installed version before consequential use.
The qualification declaration separates changed-domain checks from carried-forward numerical and installer evidence. Read the release record for exact candidate, platform and artifact results: local Windows tests do not alone establish final multi-platform acceptance or a full frozen-installer GUI lifecycle. Minimum dependencies, arbitrary large-volume memory/performance and acquired-microscopy accuracy are not implied. Existing GPU admission remains unchanged.
This alpha does not add AI pipeline generation, predictive/appearance tracking, lineage inference, deformable registration or GPU detection/tracking.