PixInsight MLDenoise Technology Preview for macOS ARM64 Released

Pleiades Astrophoto has officially announced the release of a new machine learning tool designed specifically for the noise reduction of astronomical deep sky images. This software release represents an initial preview of the next generation of machine learning PixInsight tools that the company plans to introduce in upcoming updates. The development roadmap for these advanced tools includes dedicated utilities for noise reduction, star removal, deconvolution, and various other complex tasks where specialized neural networks can assist. By integrating these neural networks into their software, the developers aim to help users solve intractable or ill posed problems by leveraging hardware acceleration capabilities. This particular software release is built exclusively for the macOS ARM64 platform, which is also known as Apple Silicon. It is currently available as a regular software update for users running PixInsight version 1.9.4 Lockhart, specifically requiring build 1693 or later.

The software development team, led by co founder and CEO Juan Conejero, has positioned this release as an early look into their broader development plans for the popular image processing platform. To utilize the new noise reduction tool, users are required to obtain a specific neural network database file. Users must download this specific file to any local folder on their computer file system, such as a dedicated directory located under their main home folder. Once downloaded, individuals can open the application and click the preferences button, represented by a wrench icon, to open a dialog where they can select this item as their default model file. While the current iteration of the software is fully functional, it is considered a technology preview and still lacks certain important features, such as real time preview capabilities. Despite this temporary limitation, the software operates highly efficiently with previews thanks to implemented caching techniques that optimize repeated execution. Looking ahead, the developers have confirmed plans to release a new version of the tool for Linux, macOS, and Windows operating systems in a future update. This forthcoming release is expected to include numerous improvements and expanded features, which will bring parity across the different hardware platforms supported by the company.

PixInsight MLDenoise Technology Preview for macOS ARM64 Features

When operating on Apple hardware, the application automatically utilizes graphics processing unit acceleration through the Apple Core ML infrastructure. This integration is exceptionally convenient for users because hardware acceleration functions as an integral component of the macOS operating system, requiring no additional configuration. For the best possible results, the development team highly recommends applying the noise reduction process to color calibrated linear deep sky images. Because dialing in the required machine learning parameters can involve trial and error, users are encouraged to rely on reduced previews focused on selected image regions of interest. This targeted approach allows users to rapidly fine tune the necessary machine learning settings using the implemented user interface. Once the optimal configuration is established and verified, the user can then confidently apply the computational process to the entire astronomical image as usual.

To improve workflow efficiency, the developers implemented a dynamic cache feature that stores partial results and other auxiliary data structures while users try different parameters on previews. By avoiding unnecessary duplicate calculations, this dynamic caching system greatly accelerates the task. The team notes that while the implemented dynamic cache is perfectly usable in its current state, it is still not ideal and will be improved considerably in the next major version of the software. Another feature included in this release is the integration of linear masks. The developers have implemented linear masks in several tools for a long time, including the Multiscale Linear Transform and Multiscale Median Transform. Within this environment, a linear mask functions as a local support for the noise reduction of linear images. These masks can be very efficient for the noise reduction of linear images that feature a strong correlation between the signal to noise ratio and illumination. Since this specific condition holds for most linear deep sky astronomical images, linear masks are highly useful. Users can evaluate the protection provided by a linear mask by turning on the preview mask option. When the process executes with this setting enabled, the target view is replaced with a representation of the mask that will be applied during normal execution. Following this specific preview execution, the standard screen transfer function of the target view is automatically disabled to allow the user to clearly view the generated mathematical mask. For normal execution on the same view, the user will have to re enable the screen transfer function by pressing Control and S on macOS, or F12 on Linux and Windows.

- DRAFT COPY ONLY -