AutoIntegrate v1.88 test - Auto stretching, Auto tools

AutoIntegrate
1

AutoIntegrate v1.88 test is a test version that is available in the test repository.

PixInsight: Script / AutoIntegrate
Test repository: https://ruuth.xyz/test/autointegrate/

Changes in AutoIntegrate v1.88 test are listed below

Auto image stretching

  • Image stretching has a new default option, Auto, which picks the stretching method when processing starts. Choosing a method explicitly always overrides it, and the target type no longer disables an explicit choice.
  • With target type Galaxy or Star cluster, Auto uses MultiscaleAdaptiveStretch, or Masked Stretch if MultiscaleAdaptiveStretch isn’t available.
  • With no target type or Nebula, Auto analyzes the image to decide:
    • MultiscaleAdaptiveStretch for images with a small, bright target such as a galaxy, star cluster or bright nebula.
    • Auto STF for images that look flat, such as frame-filling or dark nebulae.
  • The analysis measures image contrast after removing gradients and was tuned on about 100 test targets. It runs once per run, on the luminance image in LRGB and on the Ha image in narrowband, so all images get the same method.
  • It can be turned off with Use in Auto in Postprocessing / Stretching. The Sky analysis button runs the analysis on the preview image.

Tool selection

  • Settings / Tools uses radio buttons, so only one tool per processing step can be selected.
  • The first option, Auto, picks the best available tool when processing starts and prefers AI-based tools.
  • Tool settings moved into the matching processing sections. A new Sharpening and deconvolution section holds the BlurXTerminator and GraXpert deconvolution settings. The Tools tab now holds only the GraXpert path and the MLDenoise model.

Fast mode

  • New option Subset, now the default. It uses 10%, or at least 10, of the light files per filter, picked evenly from the whole list, at full image size.
  • Master calibration files are still made from all calibration files, and only the selected lights are calibrated and processed. This makes it quick to check a data set or try options and palettes.
  • A reference image you selected yourself is always included in the subset.
  • Options S and M still downsample images (at most 2048 pixels on the longest or shortest side). Their binning is fixed.

FastIntegration and large data sets

  • Fast SubframeSelector measures 32 files picked evenly from the whole session instead of the first 32. It no longer filters outliers, since only part of the files is measured.
  • The reference image is the best image from the middle half of the files if its SSWEIGHT is at least 80% of the best.
  • With field rotation on alt-az mounts, such as the Seestar, this leaves less low-coverage area at the image edges.
  • The integrated image now gets image type Master light.
  • The documentation has new tips for processing Seestar files.

Gradient correction

  • If the GradientCorrection process fails, for example with Insufficient data to sample gradient, ABE is used instead and a warning is printed at the end of processing.

Windows at end of processing

  • The new default Keep integrated and final images closes intermediate images, so AutoContinue starts from the integrated images.
  • The old Default option is renamed Keep all images.
  • Windows at the end of the processing option is in the Other / System settings section.
  • LowRejectionMap_ALL is renamed AutoCropInfo. AutoContinue still accepts the old name.

MLDenoise and noise reduction

  • The MLDenoise model file is set in Tools / MLDenoise, visible also in simple mode. Standalone Image Enhancements has the setting too.
  • The native MLDenoise process is tried first.
  • ACDNR is skipped also when DeepSNR is used, it was previously skipped when other AI tools were selected.
  • StarNet2 uses its unscreen option.

Other improvements

  • The Preview AutoSTF option is renamed AutoSTF during processing. A new option, AutoSTF when browsing, controls stretching of file previews.
  • Interface settings that are saved on exit are grouped into a Preferences box with a Save button.
  • AutoContinue, and a normal run, stop right away if a final image such as AutoRGB is already open, instead of failing at the end.
  • A clear error is shown if a setup file has no files.
  • New command line options: max_lights= limits light files per filter, and force_auto_stretching uses Autostretching.
  • ImageIntegration buffer size and stack size are now configurable.

Bug fixes

  • Fast mode S and M binning.
  • A user-selected reference image was lost when only a subset of light files was used.
  • A setup file with no files crashed with a null error.
  • The flowchart showed the MultiscaleAdaptiveStretch background step in the wrong order.
  • RGB stars deconvolution now respects No sharpening.
  • Dialog layout fixes, the Expert radio button alignment, and the manual filter files checkbox.