Release 1.1.4: replace noisy preview modes with patch denoising and fixed stretch
All checks were successful
Release AstroFrame / packages (push) Successful in 59s

This commit is contained in:
Elijah 2026-10-04 14:27:04 -07:00
parent d3362c8341
commit b8ffe97c46
17 changed files with 143 additions and 270 deletions

View file

@ -57,11 +57,11 @@ Version 1.1.0 selects up to 16 workers from CPU capacity, leaving two logical pr
Small JSON metadata stores decisions, notes, evidence and compact masks. Binary versioned caches hold stars, sky maps and forced photometry, validated against source size and modification time and reference identity. A new engine version invalidates analysis without losing human labels. Repeated analysis uses cached measurements and forced apertures when valid. Project writes replace a temporary file atomically. Cancellation leaves analysis incomplete and export disabled.
Display caches have a separate version. They contain compressed, linear, area-averaged thumbnails rather than a permanently stretched byte image. **Frame view** retains its noise-aware black point and adjustable median brightness. **Nebula view** applies a 3×3 median and four separable binomial smoothing passes at the existing thumbnail dimensions, suppressing small stars and noise while retaining faint filaments. Its black point also includes the lower background percentile, and its median is displayed slightly brighter. **Asinh view** applies an adjustable inverse-hyperbolic-sine transfer to this same smoothed plane, with a brightness-adjusted gamma fallback when its normalized median already exceeds the requested brightness.
Display caches have a separate version. They contain compressed, linear, area-averaged thumbnails rather than a permanently stretched byte image. **Frame view** retains its noise-aware black point and adjustable median brightness. **Denoised** is the only alternative. It performs non-local means on the linear thumbnail using 5×5 patches, a 13×13 search window, reflected borders, noise-bias subtraction and a filter strength of 0.8 times the estimated noise sigma. Its black point, white point and normalized median come from the original thumbnail; they are never recomputed from the denoised result. This prevents the stretch from amplifying residual noise after denoising. It adds 0.07 to the requested display median, matching the approved samples' default brightness of 0.25 at slider value 0.18.
**Color nebula** reads recognized RGGB, GRBG, GBRG or BGGR Bayer layouts and their offsets through a separate streaming display reader. It bins linear RGB superpixels, uses the same diffuse luminance preview, neutralizes each channel's median sky, smooths chroma and suppresses departures consistent with noise. A linked channel scale limits saturation and preserves neutral black and white highlights. Color is approximate and uncalibrated; no color is invented for monochrome or unknown Bayer layouts, which show the grayscale nebula preview with an explanatory label. Color has a separate fingerprinted linear cache generated only on demand. Neither display reader changes the monochrome analysis path, classifications, or original FITS files.
The previous Nebula, Asinh and Color nebula modes are removed. The denoised result stays in the selected viewer's memory and is computed only when requested. It needs no new disk cache or FITS reader and works on monochrome and Bayer-derived grayscale thumbnails alike. Existing linear caches remain compatible, and existing color caches are simply unused. The analysis path, classifications and original FITS files are unaffected.
Prepared display statistics are cached in the viewer so slider changes remain responsive. Disk loading and first-time thumbnail creation run off the UI thread, with cancellation and a single preview-loading gate preventing queued FITS reads during rapid navigation. Both comparison panes use the selected mode and shared zoom/pan. Old projects regenerate only requested displays on demand. Single-frame previews cannot recover signal absent from the exposure and can remain noisy; these are human-readable viewing aids, not stacked images.
Prepared display statistics and the denoised plane are retained in the viewer so slider changes remain responsive. Loading and denoising run off the UI thread, with cancellation, atomic memory reservations and a single preview-loading gate preventing queued work during rapid navigation. The original Frame preview remains available while denoising runs. Both comparison panes use the selected mode and shared zoom/pan. Old projects regenerate only requested displays on demand. Denoising is a viewing aid and may soften weak detail; Frame view and source-star inspection remain available for judging unfiltered morphology.
The preview is not an input to classification. Source-star inspection reads two originals, maps identical stars through their stored alignment, and displays enlarged source superpixels without interpolation. Crops are independently normalized to judge morphology. See [Siril's screen-transfer discussion](https://siril.readthedocs.io/en/latest/processing/stretching.html) for the distinction between display stretching and linear image data.

View file

@ -2,7 +2,7 @@
The owner selected the Graphite Studio palette and Plum Atlas layout from the standalone concepts. The application combines their charcoal backgrounds, silver accents, restrained decision colors, compact summaries and three-column workspace. The exposure queue is on the left, the largest column holds the preview, and the right column contains one selectable quality graph above permanently accessible evidence and measurements. A compact selector above the workspace switches sessions, including All sessions. Filtering, sorting, graph selection, keyboard review, saved overrides, undo and accepted-frame export remain available.
Frame/Nebula view, cloud overlay and nearby comparison sit above the preview. Preview stretch has a dedicated 48-pixel row below it and a visible silver thumb. The single verdict/time caption, filename and decision buttons follow the stretch. Long evidence lines present a concise first clause, with their complete engine-generated message available on hover. The evidence panel scrolls when needed; no measurements or saved explanations are discarded.
The Frame view / Denoised dropdown, cloud overlay and nearby comparison sit above the preview. Denoised uses the approved patch-based noise reduction with the original thumbnail's stretch parameters. Nebula, Asinh and Color nebula options are removed. Preview stretch has a dedicated 48-pixel row below it and a visible silver thumb. The single verdict/time caption, filename and decision buttons follow the stretch. Long evidence lines present a concise first clause, with their complete engine-generated message available on hover. The evidence panel scrolls when needed; no measurements or saved explanations are discarded.
The header now has a plain graphite background. Its height is reduced from 100 to 74 logical pixels: a 36-pixel native caption row and a 38-pixel project/action row. The project name and light/session counts sit on one line. Removing the photographic background also removes its display-bitmap allocation; the original asset remains preserved with the standalone design concepts.

View file

@ -0,0 +1,9 @@
# AstroFrame 1.1.4
- Keep the preview dropdown with just **Frame view** and **Denoised**. Remove Nebula view, Asinh view and Color nebula.
- Add the approved patch-based denoising alternative. Keep the stretch parameters derived from the original thumbnail so residual noise is not amplified into blotches. The denoised preview is slightly brighter by default and still supports the Preview stretch slider.
- Apply the selected mode to both comparison panes, preserving zoom, pan and overlays. Compute denoising in the background only when requested, reuse it for brightness changes, and cancel obsolete work when switching exposures.
Frame view's rendering and existing linear caches are unchanged. Display processing never changes original FITS files, quality measurements, classifications or manual decisions.
Validation: 95 engine, 23 updater, 10 mocked publisher and 65 isolated Windows interaction checks, with rendering at 96 and 144 DPI. Covers noise reduction, retained stretch parameters, stars and diffuse signal, constant fields, reflected borders, input preservation and cancellation. Across 11 exposures from all seven sample targets, the production renderer matches the approved samples within one gray level out of 255; Frame outputs match their previous pixels exactly. Verification uses workspace fixtures and leaves the installed app untouched. The server workflow builds Windows and Linux packages and checks the packaged Linux settings page.

View file

@ -127,7 +127,7 @@ For preview and processing-status changes, build the source and run the isolated
./tools/Verify-Preview.ps1 -FixtureProject 'path/to/saved-project' -Output artifacts/preview-verification
```
This checks comparison overlays, zoom/pan, all four preview modes in both panes, Bayer color loading across frame changes, monochrome fallback, processing-memory visibility, elongation crops and guide toggling, absent comparisons, loading cancellation, and native window controls. It renders normal and minimum-size controls at 96 and 144 DPI. Inspect the PNG files in the output's `qa` directory. It launches only the source build in diagnostic mode, which bypasses installed settings and updates. Source `--capture` diagnostics accept `--preview=0` (Frame), `1` (Nebula), `2` (Asinh) or `3` (Color nebula) for real-exposure visual checks against copied workspace projects.
This checks comparison overlays, zoom/pan, Frame and Denoised in both panes, on-demand denoising, rapid frame/mode changes, monochrome sources, processing-memory visibility, elongation crops and guide toggling, absent comparisons, loading cancellation, and native window controls. It renders normal and minimum-size controls at 96 and 144 DPI. Inspect the PNG files in the output's `qa` directory. It launches only the source build in diagnostic mode, which bypasses installed settings and updates. Source `--capture` diagnostics accept `--preview=0` (Frame) or `--preview=1` (Denoised) for real-exposure visual checks against copied workspace projects.
`./tools/Verify-ReleaseAutomation.ps1` exercises version/tag mismatch, upload failure, draft resumption, conflicting assets, immutable published releases and package integrity with a mocked release API. It never publishes a real release.