Menu: Smart Tools - setiastro/setiastrosuitepro GitHub Wiki

Smart Tools Menu

AI-assisted tools for optical correction, deconvolution/denoise/super-res, gradient cleanup, and star handling. Items marked (dialog) open a tool window; items marked (integration) launch the external engine and re-import the result. All tools create a history step (Undo/Redo supported) and can be run headlessly via Shortcuts Canvas → Alt-drag onto a view.

Riccardo’s Aberration Remover (dialog)

Corrects lens/telescope aberrations (coma/astigmatism/tilt & slight curvature) while preserving stars.

What it does

  • Estimates a spatial PSF field from detected stars.
  • Solves a low-order vector warp that recenters/rounds the PSF across the frame.
  • Optionally trims edges to avoid extrapolation artifacts.

Inputs / Outputs

  • Input: mono or RGB (linear or stretched).
  • Output: corrected image (new view or replace active).

Controls (typical)

  • Patch size, overlap, border (px), auto-GPU.
  • Strength (how aggressive the field correction is).
  • Keep edges / crop edges.

Tips

  • Run after star alignment and before strong sharpening.
  • If corners still show residual elongation, increase patch overlap; if small stars look over-tightened, lower strength a notch.

Cosmic Clarity (dialog)

Deep-learning suite with three modes: Sharpen, Denoise, and Super-Resolution. Operates on linear or gently stretched data; prefers clean, registered images.

A) Sharpen

  • Purpose: Recover detail with star-aware deconvolution; preserves color and avoids ringing.

  • Key options:

    • Sharpening mode: Stars / Non-stellar / Both
    • Auto PSF (estimates PSF per-image)
    • Amount controls for stellar & non-stellar structures
    • Optional separate-channels processing

  • Output: Replace or create a new view (configurable).

When to use: After background/white-balance, before heavy contrast curves. Works well with a mild mask on high-SNR areas.

B) Denoise

  • Purpose: Multi-scale noise reduction that respects structure and color.
  • Key options: Luma/Color denoise amounts, “full vs. detail-preserving” mode.
  • Tip: Do a light pass early (linear) and a second gentle pass near the end.

C) Super-Resolution

  • Purpose: Upscale by ×2 (or configured factor) with detail synthesis that’s star-safe.
  • Tip: Upscale before final sharpening to keep stars round.

General notes for Cosmic Clarity

  • GPU used automatically when available (fast).
  • Works on mono and RGB.
  • Create-new-view toggle is handy for A/B compare.
  • If halos appear after heavy sharpening, follow with Halo-B-Gon or a soft morphological close on stars.

Cosmic Clarity — Satellite Trail Remover (dialog)

Detects and suppresses linear satellite/airplane trails.

How it works

  • Builds a robust model of trails using a Hough-style voting + confidence map.
  • Inpaints across detected lines in a structure-aware fashion.

Controls

  • Sensitivity (increase if faint trails remain, decrease if faint filaments get flagged).
  • Line width estimate (px).

Tips

  • Best on linear (unstretched) data.
  • If residuals remain, run a second pass with slightly higher width.

GraXpert (integration)

Background/gradient removal and (optional) denoise via the GraXpert engine.

Gradient Removal

  • Modes: Classic sample-grid, RBF, or AI-based.
  • Flow: Send active view → set samples/mask in GraXpert → run → result re-imported as a new view (or replace).
  • Tips: Protect nebulae/galaxies with the sampling mask; check background neutrality afterward.

GraXpert Denoise

  • Balanced spatial/chroma denoise that works well after gradient removal and before sharpening.

Notes

  • Works on mono/RGB; linear preferred.
  • Project Save will remember that a GraXpert pass was used (via metadata).

Star Removal (dialog / integration)

Two engines are available; both produce Starless and (optionally) Stars-only results.

StarNet++

  • Traditional CNN model, fast and robust on many images.
  • Good default for general star removal.

Cosmic Clarity — Dark Star

  • Modern star-aware extractor trained to minimize halos and preserve tiny non-stellar knots.
  • Often cleaner around dense cores and small starfields.

Options & Workflow

  • Linear vs. Non-linear: Works on both; linear usually yields cleaner masks.

  • Outputs:

    • Starless view (main subject)
    • Stars-only view for recombination

  • Reintegration: Use Add Stars… (blend with opacity/color tuning) or Recombine Luminance… if working L-RGB.

Tips

  • If residual “star doughnuts” appear, run a light Morphology (open) on the starless view before recombination.
  • Keep a mask handy to limit star operations to sky only when needed.

Common Behavior & Headless Use

  • All Smart Tools honor Undo/Redo and can create a new view or replace the active view (per tool).
  • Many tools support presets; store them on Shortcuts and Alt-drag a shortcut onto a view to run headless with that preset.
  • GPU is used when available (GraXpert, Cosmic Clarity, Riccardo’s tool where supported).
  • For batch processing, place shortcuts on the desktop canvas and work left→right through your pipeline.

Quick Troubleshooting

  • Over-sharpened stars / ringing: Lower stellar amount, enable Auto PSF, or follow with Halo-B-Gon.
  • Waxiness after denoise: Reduce color denoise; add a small detail-preserving (luma) denoise only.
  • Residual gradients after GraXpert: Increase sample count or switch to RBF/AI mode; re-neutralize background.
  • Star removal artifacts: Try the other engine, run on linear, or pre-denoise lightly before removal.