Blur Xterminator AI4 disaster!

[ollypenrice]

I just downloaded AI4 and it is a catastrophe for our RASA images. We get elongated stars on one corner, some of which look almost like overlapping doubles. Blur XT with the previous AI module sometimes split them into doubles. With AI4 operating, it splits them into doubles even more perfectly and splits previously acceptable corner stars elsewhere into doubles as well! I know we should fix this with collimation and tilt but it is an undertaking to fiddle with the RASA and the problem is usually cropped out in the final cut anyway.

Does anybody know how I can go back to my previous AI package?

Olly

[geeklee]

@ollypenrice There should be a "Select AI" button on the BlurXTerminator GUI.  When clicked, I see an AI2 and AI4 file - you can then double click on the AI2 file (BlurXTerminator.2.pb) to go back to that AI package.

16 minutes ago, ollypenrice said:

Blur XT with the previous AI module sometimes split them into doubles.

Frustrating as Russ discusses this exact problem (using AI2) on the BlurXTerminator article for AI4 with demos of the old and new behaviour.  It might be worth sending him a few examples you're seeing with AI4.

[Budgie1]

Have you tried playing with the slider? 
It defaults to 0, but try running the process at 1 or somewhere in between to see if that makes a difference.

[Ouroboros]

@geeklee’s suggestion sounds about right. Opinions seem mixed about this latest version. Mostly good though. The few times I’ve tried it it’s been amazingly good. Almost so good that one wonders whether moderate amounts of  tilt or similar issues issues is going to be worth worrying about. Or for that matter … do you really need that expensive astrograph?  😀

[Stuart1971]

58 minutes ago, ollypenrice said:

I just downloaded AI4 and it is a catastrophe for our RASA images. We get elongated stars on one corner, some of which look almost like overlapping doubles. Blur XT with the previous AI module sometimes split them into doubles. With AI4 operating, it splits them into doubles even more perfectly and splits previously acceptable corner stars elsewhere into doubles as well! I know we should fix this with collimation and tilt but it is an undertaking to fiddle with the RASA and the problem is usually cropped out in the final cut anyway.

Does anybody know how I can go back to my previous AI package?

Olly

You need to reduce the star sharpening down from the new default of 0.5 to 0.25 as a start, not sure why they have changed this, but I had the exact same issue on some data of mine too, splitting an elongated star into one large with a tiny one next to it, also try with full default settings apart from the star sharpening, as I said try that on 0.25 and see, if it still does it then lower it more…

[Stuart1971]

But when you get the settings correct for you, it is far superior to the older version, that is for sure….

[Miguel1983]

Not had any issues with AI4.

I don't use star sharpening i'ts too much and indeed you could end up with artifacts.

So sharpening is set to zero en check "correct only" ☑️

These are APSC corners on my RASA 11 V2

Not corrected

Corrected

Edited December 19, 2023 by Miguel1983

[Budgie1]

1 hour ago, Budgie1 said:

Have you tried playing with the slider? 
It defaults to 0, but try running the process at 1 or somewhere in between to see if that makes a difference.

Forget the above, I was thinking about GraXpert.

I think Stuart is on the right lines.

[Stuart1971]

1 hour ago, Miguel1983 said:

Not had any issues with AI4.

I don't use star sharpening i'ts too much and indeed you could end up with artifacts.

So sharpening is set to zero en check "correct only" ☑️

These are APSC corners on my RASA 11 V2

Not corrected

Corrected

You can’t have star sharpening on 0 if using “correct only’ as its greyed out, and internal defaults are used….so stars will always be altered when using correct only….

[Miguel1983]

22 minutes ago, Stuart1971 said:

You can’t have star sharpening on 0 if using “correct only’ as its greyed out, and internal defaults are used….so stars will always be altered when using correct only….

You're correct off course, I should have said I that I only check "correct only"

[Stuart1971]

12 minutes ago, Miguel1983 said:

You're correct off course, I should have said I that I only check "correct only"

So that does sharpen and correct star shapes, but leaves nebula and structures alone….
The issue with this, is not having control on the amount of correction, to the stars, I tend to use the full setting but lower the sharpen non stellar option, and reduce the new 0.5 default of star sharpening, down to 0.25, and seems to work well….your mileage may vary….

[symmetal]

I upgraded to the latest PI version, released on 3rd Dec and after running it, as usual it needs to reinstall the 3rd party addons which it did. However BXT now only has the AI4 option available. I copied the AI2 version (BlurXTerminator.2.pb) from my other PC to the C:\Program Files\PixInsight\library folder, and the header file with the same name to the C:\Program Files\PixInsight\library\PaxHeader folder. I now have the choice of which AI to choose. 🙂

With AI2 I used to run the PSFImage script to get the image PSF pixel size as that seemed to give better results overall than the BXT Automatic PSF option. With the RASA 11 I generally had sharpen stars at 0 as it tended to create artifacts if used. However I did this with the AI4 version and the result was poor with almost no corner star shape correction. Using the Automatic PSF it did correct the corner stars perfectly as well as sharpening them, (if selected). Possibly AI4 recomputes the PSF for different areas of the image so I'll use that from now on. The AI4 sharpen stars doesn't create the artifacts the AI2 version did either.

@ollypenrice If you need the AI2 files send me a PM. 🙂

Alan

Edited December 19, 2023 by symmetal

[Miguel1983]

59 minutes ago, Stuart1971 said:

So that does sharpen and correct star shapes, but leaves nebula and structures alone….
The issue with this, is not having control on the amount of correction, to the stars, I tend to use the full setting but lower the sharpen non stellar option, and reduce the new 0.5 default of star sharpening, down to 0.25, and seems to work well….your mileage may vary….

it is my experience so far that "correct only" makes star shapes better but dus not do any sharpening, but that can also just be my misguided perception 😅

[Stuart1971]

2 hours ago, Miguel1983 said:

it is my experience so far that "correct only" makes star shapes better but dus not do any sharpening, but that can also just be my misguided perception 😅

It is only subtle I think, and does depend on the quality of the stars to begin with, good data in better data out and less adjustment needed….👍🏻

[Budgie1]

4 hours ago, symmetal said:

With AI2 I used to run the PSFImage script to get the image PSF pixel size as that seemed to give better results overall than the BXT Automatic PSF option. With the RASA 11 I generally had sharpen stars at 0 as it tended to create artifacts if used. However I did this with the AI4 version and the result was poor with almost no corner star shape correction. Using the Automatic PSF it did correct the corner stars perfectly as well as sharpening them, (if selected). Possibly AI4 recomputes the PSF for different areas of the image so I'll use that from now on.

That's interesting, Alan. I've always used PSFImage to get the correct PSF figure before running BXT on nebula, only using the auto option on galaxies.

So it'll be interesting to see the comparison between auto & manual in AI4 for me.

[Scott Badger]

FWIW, my first try with AI4 resulted in ringing around some stars which didn’t occur with AI2 at the same settings, but there was no ringing after running Correct Only first.

Cheers,

Scott