Skip to content

v0.2.1-alpha

Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 20 Oct 06:05
· 4 commits to main since this release

OPIPP 0.2.1-alpha (Open PIPP) - Windows x64

Self-contained alpha build of the managed PIPP continuation. It was built and
tested, packaged, and smoke-tested from the extracted zip by the tag-triggered
release workflow before this release was created.

Artifact OPIPP-0.2.1-alpha-win-x64.zip
Size 53539535 bytes
SHA-256 d28d4be9f328d2d3ca469b2e485501e61905cb87b473199ecfec7aa76df621a9
Tag v0.2.1-alpha
Commit 15d713d
Workflow run https://github.com/openpipp/opipp/actions/runs/39378137361

Package contents

  • Opipp.App.exe - the GUI, fully managed, no pipp_dll.dll
  • Opipp.Cli.exe - the command-line engine, drop-in for pipp_core.exe
  • README-ALPHA.txt - quick start, status and licence notes

Status and licence

ALPHA: a working preview, not an official PIPP release, and not affiliated
with Chris Garry or the PIPP project. AI was used to build this. GPLv3 or
later for this repository's code and artwork; not a licence to PIPP. See
NOTICE in the repository.

What changed in 0.2.1

A correctness patch for final-image integration. Every fix closes a way to get a wrong image, destroy
someone else's file, or silently ignore a configured resource limit.

Raw CFA frames are no longer averaged across colour sites. Integration compared Bayer tags before
registration, but an odd-pixel shift moves the 2×2 mosaic to another phase, and bilinear sampling on a
raw mosaic blends four differently coloured sites. Each contributor's effective phase is now computed
after placement and after the output crop: the output is tagged at its actual origin, contributors that
disagree are rejected before any sample changes, and a fractional offset on a raw mosaic is refused
outright with a message pointing at debayering. (A single frame shifted by −1 is now correctly tagged
GRBG; two RGGB frames placed one pixel apart are rejected instead of averaged.)

Integration no longer reloads the retained capture. The pipeline used to materialise every selected
frame into a list before integrating, defeating the retention budget; median then spilled them again
while the full set stayed reachable. Frames are now read from the retention store one at a time (and
re-read per median band) with no frame held across calls. A 300-frame spilled selection peaks at about
2 MB of live objects instead of ~19 MB.

Staging files are owned, not guessed. The predictable <name>.integrating<ext> path could truncate
and then move away an unrelated existing file. Staging now uses a unique GUID name created with
CreateNew, and cleanup deletes only that exact path. A preseeded .integrating sentinel survives
success, failure and cancellation.

The area of interest is applied after registration. Cropping before registration threw away source
pixels needed inside the requested aligned crop (a 10×10 AOI came out 7×9 with 63 valid samples). The
AOI is now a request on the delivered canvas and is applied to the integrated result, giving the
requested 10×10 and 100 valid samples; a request that misses the overlap fails with both rectangles named.

Median honours its in-memory threshold. Every median selection was forced through the disk path,
ignoring MaxRetainedMedianFrames; with the default budget a 44-frame Full-HD RGB stack failed although
it is below the cap. Median now routes in memory when the count and a declared byte estimate fit, and to
the disk-backed path otherwise. Both routes produce identical pixels; the chosen route and estimate are
reported on the result.

GUI integration settings are part of the option lifecycle. Enabled, method, coverage and subpixel
are written to .pof files, restored on load (old files get explicit defaults), cleared by Reset
Options and restored by Undo Reset. Previously a reset run still emitted -stack=sum.

Corrections to earlier claims

  • "One file per run" was wrong: integration produces one file per batch input, or one per joined unit.
  • "Median stays in memory up to 64 frames" was not true in production before this patch; it now is,
    subject to the byte estimate, with the disk path above it.
  • A run where an input fails still publishes the accepted frames' integration as a labelled partial
    result with exit 1; final integration does not require every selected file to decode.
  • A hard kill during output staging can still leave one uniquely named .integrating orphan, because
    the engine cannot run cleanup when the GUI kills it. It is never mistaken for a finished image, and
    the GUI deletes only files it owns rather than sweeping the directory.

Verification

  • 1,319 tests, 0 failures, nine projects.
  • Parity unchanged: 27 compared, 27 matched, 2 deliberate differences, 0 gaps.
  • Reproduced against the built CLI: two RGGB frames whose centring places them one pixel apart now exit
    1 with the phase conflict named and write nothing; a static odd-shift case succeeds with the tag
    computed at the output origin; the 40×30 AOI case returns a 10×10, all-200 image; a preseeded
    a.integrating.ser survives an integrating run unchanged.

Known limits

  • No sigma-clipped or robust rejection; every contributing frame counts.
  • Integration output is quantised to the input depth (double intermediate, integer delivery); no float
    FITS output yet.
  • Raw mosaics reject fractional registration and phase-changing shifts; debayer to colour first.
  • Subpixel translation only: no rotation, drizzle or local warping.
  • Median above the in-memory caps is exact but slower, and scratch use still grows with contributor
    count at a fixed band height.

OPIPP 0.2.1-alpha (Open PIPP) - Windows x64

An independent, fully managed continuation of PIPP (Planetary Imaging
PreProcessor). ALPHA: a working preview, not an official PIPP release.

NOT AFFILIATED WITH PIPP. Not affiliated with, endorsed by or supported by Chris
Garry or the PIPP project. New C# code, no original binaries redistributed.
Published in the belief that this is fair use.

AI was used to build this and will keep being used. Behaviour claims are backed
by tests or parity cases; unverified items are listed as known gaps.

What is in this folder
Opipp.App.exe the GUI (managed; no native pipp_dll.dll needed)
Opipp.Cli.exe the command-line engine, drop-in for pipp_core.exe

Quick start
GUI: double-click Opipp.App.exe
CLI: Opipp.Cli.exe -help
Opipp.Cli.exe -tif -od=C:\out capture.ser

Status
Built automatically from tag v0.2.1-alpha by the release workflow after the
full test suite passed. This exact zip was extracted and smoke-tested
(a generated SER file converted to TIFF) before the release was published.
Parity against the original PIPP engine is not run in CI because it needs
the original binaries, which are not redistributable; see docs/parity.md
and docs/status.md at the tagged commit for the accounting and known gaps.
Workflow run: https://github.com/openpipp/opipp/actions/runs/39378137361

Licence
GPLv3 or later for this repository's code and artwork. Not a licence to PIPP.

Requires Windows 10/11 x64. Self-contained: no .NET installation needed.

Checksum

d28d4be9f328d2d3ca469b2e485501e61905cb87b473199ecfec7aa76df621a9  OPIPP-0.2.1-alpha-win-x64.zip