Skip to content

VIPP 0.14.0a1

VIPP 0.14.0a1 makes scientific image selection explicit and durable. The same selected item, reader evidence, source revision, axes, metadata, and effective per-sample parameters now travel through interactive calculation, batch, generated execution, replay, export, checkpoints, manifests, and provenance.

This remains alpha software. Keep original data and workflows, test representative sources through the intended reader, and review important outputs before using them for scientific conclusions or publication.

Highlights

A durable SourceItem

Image Source now records a frozen SourceItem v1 identity for the logical image selected inside a file or store. That identity separates:

  • the stable item selector;
  • the observed file, directory, or companion-file revision; and
  • the reader/backend and normalized shape, axes, and metadata that resolved it.

Stable keys replace order-dependent series selection where a reader can provide them. Changed bytes, missing companion files, ambiguous legacy indices, or an unexpected reader topology stop visibly rather than selecting a different image. Portable provenance does not add exact local paths.

Workflow schema 5 and batch config/manifest schema 4 write this representation. Existing workflow schemas 3 and 4 and batch versions 1 through 3 retain explicit migration paths; review and save a copy before replacing an important original.

More truthful microscope-reader boundaries

The frozen release corpus exercises the claimed ND2, LIF, CZI, LSM, Olympus OIR/OIB/OIF/VSI, and Imaris IMS routes. Reader inspection and full read now use one normalized metadata contract, multifile VSI/ETS and OIF trees participate in source revision identity, and missing optional readers or Java/Bio-Formats prerequisites produce actionable failures.

This does not make every reader lazy or equivalent. Native LIF, CZI, OIR, OIB, and LSM pixel access remains eager. In the qualified Leica PR2729 example, liffile exposes one combined item while Bio-Formats exposes four; VIPP pins the reviewed backend and refuses an unexplained topology change.

A lower-level preview, full-resolution analysis

For a local multiscale OME-Zarr 0.4/0.5 image or label, Image Source can display a sliced lower level labelled Preview level N - analysis remains full resolution. Its dynamic preview-level chooser lists the levels actually declared by that source. Requested T/Z/C positions and the Y/X region are selected before the preview is computed, and label previews retain label semantics.

The scientific graph still receives level 0. The lower level is presentation only: it does not replace the canonical SourceItem, scientific cache, batch input, generated execution, or output provenance. A single-level source says that no lower-level preview exists.

Reviewed values for individual batch samples

After a fresh batch plan resolves stable SourceItems, Per-sample parameters (optional) presents eligible authored numeric parameters as columns. Enter only exceptions; a blank cell visibly inherits the workflow value. Preview, batch, generated execution, checkpoints, manifests, and provenance share the same detached effective workflow without changing the authored graph.

Duplicate, stale, missing, multiply matched, or invalid typed overrides stop preflight. This release does not support expressions, filename rules, source selectors, topology changes, or CSV import/export for per-sample values.

A retained Batch workspace that explains its work

Reopening a workflow with an attached Batch workspace starts metadata-only sample discovery in the background and restores exact per-sample overrides. A compact status and activity indicator sits beside the Batch toolbar; detailed per-item and per-operation progress remains in the run section. No representative pixels are calculated until requested.

Ordinary output collisions now default to Ask before overwrite (recommended). The dialog lists the exact existing files and asks for one-run consent; Cancel leaves them untouched. Duplicate destinations, source overlaps, explicitly protected outputs, and headless ambiguity remain hard errors.

Understandable Windows setup activity

The setup review separates approximate capacity from the conservative disk gates it enforces:

Track Approximate download Approximate installed Approximate peak working space Enforced installation-drive minimum Enforced temp/records-drive minimum
CPU 250 MiB 1.5 GiB 2.5 GiB 5 GiB 1 GiB
CUDA 13 1.5 GiB 5 GiB 7 GiB 15 GiB 5 GiB

These are disk-storage figures, not GPU-memory (VRAM) comparisons. Actual downloads vary with the package index and cache. During setup the current phase, elapsed time, quiet-operation heartbeat, latest concrete activity, and setup-log path remain available. Progress is determinate only when the dependency tool reports a trustworthy byte total; a long quiet operation is not relabelled as a failure.

Compatibility and migration

  • VIPP accepts workflow schemas 3, 4, and 5 and writes schema 5. Older valid workflows acquire SourceItems when their sources are resolved.
  • Batch configs and manifests are schema 4. Older configs have no per-sample overrides; version 3 retains its source declarations and acquires SourceItems when resolved.
  • Workflow JSON does not contain source pixels or cached scientific results. Recalculate and validate after upgrading.
  • Generated Python remains pinned to the VIPP version that created it; regenerate exports after upgrading.
  • Preserve an important original before migration, then verify the selected item, reader/backend, axes, calibration, and decisive outputs.

Remaining limitations

  • Scientific graph execution still materializes the complete selected level-0 image. The lower OME-Zarr level is display-only.
  • Presentation preview is limited to local OME-Zarr 0.4/0.5. Remote stores, IMS pyramid preview, HCS plate/well/field traversal, and arbitrary operation-level lazy or chunked execution are not included.
  • Native LIF, CZI, OIR, OIB, and LSM pixel reads remain eager. VIPP can preflight estimated decoded memory but cannot invent chunk-level progress for a monolithic reader call.
  • VSI/IMS through Bio-Formats requires the optional reader stack, Java, and required codecs; these are not bundled in the base package.
  • Per-sample overrides are numeric scalar parameters only.
  • Source identity and metadata checks cannot prove acquisition truth, biological registration, or the scientific validity of a pipeline.

Qualification boundary

Focused core, reader-contract, schema/migration, preview, override, and installer-presentation checks cover the changed behavior. Exact-tag package and installer artifacts are regenerated for this release. The unchanged full GPU catalogue and installer transactional lifecycle carry forward from their recorded 0.13 evidence; this does not turn unperformed reader, GPU, fresh-machine, or biological validation into a pass.

The application release and PyPI project page are the canonical distribution locations. The GitHub release carries the exact wheel, source archive, installer, checksum, release manifest, and notices for the immutable tag.

Release verification

Gate or artifact Verified result
Annotated immutable v0.14.0a1 tag e16ca87161ec7b1041a5e98c5a2bf786b11a1ec8
Exact-main CI Passed: run 32975633514
Exact-main Windows installer build Passed: run 32975659650
Strict public corpus v4 Passed: manifest SHA-256 3365b4cec7a220f6399f3c030eb3ac751581bde730fac60f01c9fca3b823714d
Public GitHub alpha prerelease Passed: exact qualified assets
PyPI publication Passed: workflow 32982120913; downloaded wheel and source archive matched the exact release bytes
napari_vipp-0.14.0a1-py3-none-any.whl ebc46ac87d90b1b5a82821920b13f75e75f306b3301882745969e7081236be70
napari_vipp-0.14.0a1.tar.gz bad176628678d00b6cd58965c29022a55c6c7d40f414a259f2aa6e5bb70fa53d
VIPP-Setup-0.14.0a1-Windows-x86_64-UNSIGNED.exe 24ca7e8a0ea80bb21f849a30a5946e83e9e55be785047a5164305fab84d9aeea
VIPP-Setup-0.14.0a1-Windows-x86_64-UNSIGNED-release.json 066726da2b5f6a0ec130f09d005f696abdca5d5c72ddbbf211b6aae71713ca66
VIPP-Setup-0.14.0a1-Windows-x86_64-UNSIGNED-THIRD-PARTY-NOTICES.txt 95ca668d0977be347ab39af820f97652ed3565ed7ea4218205bb3a716a847e2d
SHA256SUMS-Windows-0.14.0a1.txt bf2ad89fb5e78b7f068d7ce7ee694260304535f85fe8a439d4e7fd8bd3e50032

This table records application qualification and publication only; it does not claim a documentation deployment result.

Continue reading