VIPP 0.15.0a3¶
Release date: 2026-09-09.
VIPP 0.15.0a3 brings reviewed reproducibility packages, original-input checks and verified interrupted-batch recovery to VIPP, alongside object QC and desktop workflow improvements.
Alpha: preserve the original analysis
Preserve original data, workflows, environments and decisive outputs. Recheck consequential results after upgrading. Qualification applies only to the recorded software, hardware, inputs and tested behavior.
Review, share and repeat an analysis¶
Export a reproducibility package from the current recipe or a recorded batch's run report. Review the offline report, exact files and sharing omissions before saving a local ZIP. Recorded packages use the archived run's workflow/settings, not the graph edited later.
The package includes portable workflow/batch settings, explanatory notes, advanced Python runners, software records and redacted evidence. It excludes raw images, outputs and previews; preparation/export do not upload data or install software. Folder hiding and optional filename anonymisation do not guarantee anonymity: review retained notes and scientific labels before sharing.
Open the extracted workflow in VIPP and choose Reproduce original run or Use workflow on new data. Relink inputs, choose a new output folder, then Check, review and Run. Reproduction compares original input contents and image selections; a different or unknown VIPP version needs explicit acknowledgement. Clearer check details, corrected Run handoff and completed-run status retain that distinction. A new run still requires a new check.
Matching inputs and VIPP versions do not prove identical outputs, dependencies, hardware or scientific validity. Supply data/results separately and compare consequential results independently. Packages are not locked environments or original resume receipts.
Recover a saved batch run¶
Resume saved run…
continues the archived workflow/settings without changing the open graph.
Newly generated batch runners also accept --resume MANIFEST. VIPP reuses only
whole completed items with verified input/output bytes, settings, runtime and
original evidence; the report distinguishes verified reuse from new writes.
Ordinary and resumed runs share a destination lock.
Keep original manifests and sidecars. Earlier manifests, moved runs, partial-item recovery and cross-version recovery are not supported by this first resume contract. A reproducibility ZIP cannot replace those records.
Inspect objects and handle workflows¶
- Find Label Boundaries creates a Boolean QC mask with explicit inside, outside or both-side placement, 2D/3D processing and Face/Full connectivity. Grid, calibration and upstream labels are preserved; it is a CPU operation, not a mesh or an object-ID output. See label cleanup.
- Filter result shows exact current input, kept and removed object counts for supported label/mask filters without changing calculated pixels. Stale, bypassed and uncached results do not show misleading counts.
- Drag workflow JSON onto the graph, tabs or inspector to open a new tab while preserving existing work. Recorded workflows retain the reproduction choice.
- Try Separate Overlapping Objects. The mesh example is consolidated into one authored Mesh Objects, Colours & Refinement workflow; the duplicate tuned entry is removed.
- Installed desktop launches gain VIPP branding and a title-bar-free startup window: drag its background to move it or use its small top-right minimize button. It does not stay above other windows. Plugin/manual launches retain napari branding. Initial workflow-dock space and the Batch/Display toolbar separation are clearer.
GPU execution and compatibility¶
Broader RL/RL-TV GPU ranges admit more authored options with numerical-difference advisories in Compute and execution provenance version 2. Execution permission is separate from CPU-parity or restoration-quality validation. Finite float32 input, axes, PSF mass, memory, environment and cancellation checks remain enforced.
Workflow and batch-config schema numbers remain 6. Manifest schema 6 adds recovery snapshots and run lineage; a2 writes manifest schema 5. A shared schema number does not make new operations or reproduction references readable by older runtimes. Preserve old records and regenerate Python exports for the exact installed version.
The macOS installer constrains its bundled YAML dependency to conda's compatible range. Installed desktop launchers also preserve authenticated startup when a generated token begins with a hyphen.
Installation and qualification¶
Use the canonical 0.15.0a3 GitHub release for exact assets, checksums and qualification evidence, or PyPI 0.15.0a3 for an exact package pin. Verify the matching checksum before opening an installer. Windows alpha installers are explicitly unsigned; Apple Silicon and Intel macOS packages are separate, unsigned, unnotarized and CPU-only. Follow installation.
For a manual CPU environment:
On macOS use napari[pyside6] instead of napari[pyqt6].
The application qualification declaration separates current checks from carried-forward evidence. Package installation, desktop acceptance and scientific output comparison establish different facts. User acceptance of a workflow is not independent output-equivalence evidence.