StarTools

Hi all,

With development ongoing on 1.9 and the holidays fast approaching, I wanted to at least provide a preview of some of the upcoming enhancements and quality of life improvements you can expect in the new year.

Most R&D efforts have, once again, gone into SV Decon, which should now be easier to use and yield improved results. Sampled decon in particular should show a marked improvement in SV Decon's ability to coalesce stars back into point lights, precisely as an "ideal"/theoretical solution should appear. This, for example, greatly reduces the need for things like the Shrink module.

Absent from this release is the considerable amount of work that has been going into multi-language support (this release is currently still English-only).

CHANGELOG as follows;

1.9.529 Internal Alpha 1
* Added multi-language support engine (updating of interface and translation ongoing)
* Improved Correlation Filtering in Wipe
* Added new color compositing modes in NBAccent module
* Exposure times now automatically detected and prepopulated in Compose module
* Exposure times now automatically scale by factor of 10x if >12h detected in Compose module
* Reversing sRGB stretch now tries to undo color balancing
* Support for James Web Space Telescope datasets, with automatic artifact clean-up
* Improvements to SV Decon handling of highlights and dynamic range extension and reduction of artifacts around over-exposing stars
* Removal of linearity parameter and replacement with highlight restore parameter in SV Decon
* Some small optimizations to SV Decon
* Addition of new primary apodization mask generator option to SV Decon (based on modified version of AltMask preset in automask generator)
* Mask is no longer required for de-ringing in SV Decon

1.9.531 Internal Alpha 2
* SV Decon no longer requires separation of samples when using built-in mask generator
* Added DSO Core preset to HDR module
* Added "Spectrum and Filters" and "Scene" selectors to Compose module
* SV Decon now automatically determines sample size (needs further validation testing)
* Made Color module (and its UI) spectrum and filter-aware
* Added Cap Purple for SHO narrowband composites to Color module
* Automatic preset and matrix selection based on spectrum and filters
* Further Improved Correlation Filtering in Wipe
* Made tweak to noise estimation in Tracking engine
* Slight improvement to Contrast module noise propagation handling
* Added new Integer binning modes to Bin module
* NBAccent retaining settings in setup
* SV Decon enhanced performance on heavily distorted PSFs (re-implementation of Spatial Distortion parameter)

1.9.533 Internal Alpha 3
* Complete removal of need for mask generation intervention by user in SV Decon
* New de-ringing algorithm with simplified controls in SV Decon
* Improved GPU acceleration in SV Decon
* Enhanced GPU throughput through optimized data persistence in GPU memory in SV Decon

1.9.534 Public Alpha 1
* Made more modules object, spectrum and filter aware
* Improved handling of noise propagation in Entropy module
* Made Entropy module real-time
* "Spectrum and Filters" DSLR setting now showing DSLR matrices in Color module
* Fixed glitch in Correlation Filter in Wipe
* Added Euler's number preset for spatial aberration parameter in SVDecon

As said, I got a fair way to go until we hit beta, but the improvements are significant enough that I thought them worth sharing.

Clear skies!

PS Many thanks to @hixx for lots of testing and German translations!

wow there's lots to like again in this release, super well done, thanks
SV Decon updates read great for my not so point like stars data

Hi Ivo, thank you for sharing this alpha version . I'm looking forward to spend some time with it and try it out.
At first sight, I'm not sure if I will ever learn that AutoDev is now OptiDev

Best regards, Dietmar.

Thanks Ivo! An early x-mas.

Plus...Opti!

Quite the list of changes. Amazing how much has improved from 6 to now...9...in relatively short time.

This will be fun when it's time to start tweaking towards beta and then final.

can’t wait!
Thanks!

Superb! Thanks for the alpha!

/stefan

Hi Ivo,
thanks for public 1.9 alpha!
I'm just wondering about the "Euler" setting in SVDecon - any mathematical background to e for Decon or could it just as well be Pi?
Clear Skies,
Jochen

hixx wrote: ↑Sat Dec 03, 2022 7:18 pm I'm just wondering about the "Euler" setting in SVDecon - any mathematical background to e for Decon or could it just as well be Pi?

Euler's number is everywhere and, for the purpose here, features heavily in anything to do with statistics and, for example a Gaussian blurring function.

It's actually not used in a particular sophisticated manner here. It's just a good starting point because, in the case of heavily deformed stars, it tends to yield a good "blurry" area around a central point that is still reminiscent of a star (e.g. a point light, tapering off in all directions due to diffraction), without becoming too blurry in any particular direction. The SVDecon module is basically told here (per iteration); as long as the new pixel locations ("changes") fall within this blurry area, we're good. (you should notice stars coalescing and changing more dramatically in "roundness", however never becoming more resolved than the "blurry" area).

hixx wrote: ↑Sat Dec 03, 2022 7:18 pm I'm just wondering about the "Euler" setting in SVDecon - any mathematical background to e for Decon or could it just as well be Pi?

I tried Pi, 0, 1, sqrt(2) and some of the most important physical constants (don't know where to place the units?! ) as well, but they all have one thing in common: nothing happens. Maybe my PSF samples are not distorted enough ... ?

BTW: I'm missing the new 'highlight restore parameter in SV Decon' mentioned for 1.9.529. Is/was it needless due to following changes of SV Decon?

I had a go with my M33 dataset and SV Decon 1.9 has an obviously different impact. For example, the 'stars' (clusters or whatever it is) in NGC 604 are even more tightened than before . But now I notice dark pixels in the cores of bright stars. Maybe I used unsuitable PSF samples? I remember, there was a discussion (with Martin?) regarding this, but for me / my datasets this is new.
And I'm thinking about the impact of SV Decon on slightly larger structures than the small(est) stars. I've the impression, that this was stronger with the previous version of SV Decon, but I cannot confirm this at 100% right now. First I will try some older datasets of which I remember that there was a great impact on 'clarity' with SV Decon.

Best regards, Dietmar.

P.S. / Edit: I don't know if this is worth to discuss right now. I just wanted to share my first findings. Once again, many thanks to Ivo

Interesting, Dietmar. In a few quick tests I've actually noticed less dark-spotting inside stars than in 1.8. And boy does it pinpoint. But, at least for the data I've tried, deringing is a necessity so that stars backing things off.

Sometime soon I'll put my laptop next to the desktop and run 1.8 and 1.9 side-by-side on the same data, as I had similar thoughts. It might also be that more iterations could be had.

And yeah unsure if we would start listing things out yet from our tire-kicking when this is a holiday preview.