BlurExterminator 2.0

[gorann]

It came out today and it is just amazing. I tried it on a tilted image I had, and here is the worst corner. First image is with Blur XT1 and the second with Blur XT2. I rest my case!

Cheers, Göran

Edited December 14, 2023 by gorann

[WolfieGlos]

Holy moly, that would fix so many of my images with ease. That’s remarkable, and I’d be intrigued as to whether the stars get corrected back to their “correct location”?

One of these days I’m going to have to invest in PI.

I think they missed a trick though…surely it should be TiltExterminator 🤔

I wonder if this would work on trailed images too, such as a wind affected sub? 

[tomato]

Wow, that’s impressive, but it will no doubt revive the debate on whether these “fix it” tools are ethical.

Personally I’m fine with them.

[Clarkey]

Much as that is impressive, with all the new tools we will get to the point of loading subs and let the software (AI) process it for you. Yes, I use PI and BXT, but there is an element of guilt. Part of the challenge of AP is good data collection and processing. Maybe it is getting too easy?

[Ouroboros]

Nice example, @gorann.

Adam Block released a video on this today with Russ Croman in which they discuss BlurXterminator using AI4 applied to several images.

https://youtu.be/nLyZGzT8T5c?si=eBGeQGJKLI_bRAhH

[gorann]

4 minutes ago, tomato said:

Wow, that’s impressive, but it will no doubt revive the debate on whether these “fix it” tools are ethical.

Personally I’m fine with them.

Yes Steve, the way I found out about its release was from a discussion post on Astrobin saying:

"I just watched a video on Youtube about the new version of BlurXterminator 2 and it makes me think about our IOTD and AI process:
With the new version, the following criteria lose importance
- BlurExterminator turns eggs into stars
- BlurXterminator fixes optical problems on peripheral stars
- BlurXterminator allows you to render a stack of 15 frames with more detail than 50 frames
I was wondering if we could not require in the final for the IOTD the stack of raw luminance frames so as not to let anything slip through.. "

So it got a least one person very worried (probably someone that really got his imaging rig perfectly tuned and never had to detach a camera to shift a filter in a RASA scope and hope for the best). However, it got me excited about what BlurXT2 will do for me!

[saac]

21 minutes ago, Clarkey said:

Much as that is impressive, with all the new tools we will get to the point of loading subs and let the software (AI) process it for you. Yes, I use PI and BXT, but there is an element of guilt. Part of the challenge of AP is good data collection and processing. Maybe it is getting too easy?

Really depends on what you personally want to get out of it. When I listen to music I don't want to build the amplifier myself nor do I fully understand the electronics that sample and convert from analogue to digital. I'm not interested in that - I just want to listen to the music. So I guess it depends on what you want out of your processing activity.  How many of us actually understand what any of the processing tools are doing and to what depth of understanding.  From my own point of view I'd be happy so long as I know that my data is not being added to , in effect producing something that was not there.  With that aim in mind then I'd be happy for it to be made as simple as possible.  I can well understand though that for those who really enjoy the art of processing they may chose to use a different set of tools. 

Jim 

Edited December 14, 2023 by saac

[tomato]

As someone who started out with emulsion film and manual guiding, it is just another technological advance as far as I am concerned. I wasn't really around AP when the digital revolution got going, but was there a backlash against the principle of stacking multiple frames to improve SNR, or did folks just embrace it?

[tomato]

34 minutes ago, Clarkey said:

Much as that is impressive, with all the new tools we will get to the point of loading subs and let the software (AI) process it for you. Yes, I use PI and BXT, but there is an element of guilt. Part of the challenge of AP is good data collection and processing. Maybe it is getting too easy?

I presume you  image from your location in the North West of England, so I am sure you will agree with me that getting data from the UK, good or indifferent, is never easy.😉

[saac]

11 minutes ago, tomato said:

As someone who started out with emulsion film and manual guiding, it is just another technological advance as far as I am concerned. I wasn't really around AP when the digital revolution got going, but was there a backlash against the principle of stacking multiple frames to improve SNR, or did folks just embrace it?

Now photographing astro with wet film is not something I would have taken to - you have my respect   From a scientific view point reducing SNR in idata in such a way  as stacking does is a pretty standard routine (legitimate if you want to use such a term) . I cannot imagine why, from a technical perspective anybody could have grounds to be concerned. Too be honest, I would be more concerned that it was not being exploited  when the technology permitted it. I guarantee that professional astronomers would have welcomed the digital revolution with open arms.   

Jim 

Edited December 14, 2023 by saac

[archiebald]

Hi, first post here.

For the hobbyist, I think the "moral" choice is purely personal, however the use of AI in image processing is already creating difficulties in very prestigious general photographic competitions such as National Geographic.  Since winning one of those competitions can be the making or breaking of a professional career, there is a lot riding on the policing of "legitimate" photography.  There are numerous high profile cases over the last few years of winners being disqualified.

Maybe there should be a range of separate categories for "natural" and "processed" images with strict criteria for each, including submission of raw data.

[ONIKKINEN]

1 hour ago, archiebald said:

Hi, first post here.

For the hobbyist, I think the "moral" choice is purely personal, however the use of AI in image processing is already creating difficulties in very prestigious general photographic competitions such as National Geographic.  Since winning one of those competitions can be the making or breaking of a professional career, there is a lot riding on the policing of "legitimate" photography.  There are numerous high profile cases over the last few years of winners being disqualified.

Maybe there should be a range of separate categories for "natural" and "processed" images with strict criteria for each, including submission of raw data.

Hi, and a warm welcome to SGL!

For conventional photography, especially for contests and such i do agree that generative AI is a potential issue and one the photographer must mention was used or simply not use at all if it was against the rules. However BXT is a an AI tool for a very specific purpose and does no additive or generative work on the image and every decision it does was based on measurements from the input image, so no new information is created out of thin air like with other AI image processing tools like TopazAI.

If you would like to know more about how BlurXterminator works and how it is able to make the decisions it does, check out this interview: https://www.youtube.com/watch?v=6hkVBnYYlss&t=1s

An interview of the new AI4 version of BXT can be seen here: https://www.youtube.com/watch?v=nLyZGzT8T5c

Not sure how a distinction between a natural and a processed image could be made, since every astrophotograph must be processed before it can be presented (linear data is full black basically). How well someone manages to process their image is for the viewer to determine (and voice opinions on) in my opinion.

Edited December 15, 2023 by ONIKKINEN

[archiebald]

2 minutes ago, ONIKKINEN said:

Not sure how a distinction between a natural and a processed image could be made, since every astrophotograph must be processed before it can be presented (linear data is full black basically). How well someone manages to process their image is for the viewer to determine (and voice opinions on) in my opinion.

Thanks for the welcome,

The question then may come, what percentage of an image is based on photographic skill versus processing skill.  And, taking it to the extreme, I've never tried any of the "AI" image programs or apps, but I would not be surprised if a full Deep Space AI image could be generated in a very convincing manner with data that never passed through a lens, filter or sensor.

[ONIKKINEN]

Just now, archiebald said:

Thanks for the welcome,

The question then may come, what percentage of an image is based on photographic skill versus processing skill.  And, taking it to the extreme, I've never tried any of the "AI" image programs or apps, but I would not be surprised if a full Deep Space AI image could be generated in a very convincing manner with data that never passed through a lens, filter or sensor.

Processing is probably more than 90% of the end result, if not more. I think a person skilled in processing can come up with a better image through a kit lens and a DSLR than a complete beginner with 10k € invested in their kit. Having good data helps a lot, but there is still much more to learn in processing than there is in image capture, and i think the image capture part is something most beginners will get the hang of in a couple of years, to a point where only equipment upgrades bring a noticeable improvement.

[ONIKKINEN]

Ran BXT in correct only mode with AI2 and the new AI4 on the most cursed stack i could find at the moment, side by side below. I dont think i need to point out which image was with which AI version...

And the original uncorrected version:

In short, amazing. The new AI has also managed to merge many of the unfocused diffraction spikes (but not all it seems), which was one of the issues i had with the old AI.

[tomato]

If you get your rig sorted then data capture is pretty straightforward, although my unattended sessions rarely deliver data as good as those where I sit with the scopes throughout the session.

Like it or not, I think the time where we can let the software alone deliver an optimum image from the data supplied is fast approaching. But that’s Ok, I can choose to do that or not, just as I can choose to manually find the object or focus during the data capture. 

[saac]

3 hours ago, tomato said:

If you get your rig sorted then data capture is pretty straightforward, although my unattended sessions rarely deliver data as good as those where I sit with the scopes throughout the session.

Like it or not, I think the time where we can let the software alone deliver an optimum image from the data supplied is fast approaching. But that’s Ok, I can choose to do that or not, just as I can choose to manually find the object or focus during the data capture. 

Totally agree, I don't think we are that far off the likes of ZWO offering an ASiAir type of product that manages data capture, completes processing scripts to deliver a "finished" image. The ASiAir (as others) already offers stacking and production of master image files. Like everything else with astrophotography you can decide what level of activity you wish to engage with - nice to have the choice. 

Jim 

Edited December 15, 2023 by saac

[vlaiv]

12 hours ago, tomato said:

I wasn't really around AP when the digital revolution got going, but was there a backlash against the principle of stacking multiple frames to improve SNR, or did folks just embrace it?

For me, there is clear distinction between the two.

In case of stacking images - there is precisely defined mathematical workflow that has been proven to be correct.

In second case - it's much like giving someone a list of numbers and asking them to guess which number comes next. Sure, it is educated guess (AI has to be trained) - but it is still a guess and as such - not 100% right all the time.

When solving a mathematical equation, you don't say - here, this is solution, but I'm 87% confident it is the right one , If you apply correctly mathematical principles - you can be 100% sure you have the right solution.

Since I consider astronomical imaging to be more than producing pretty pictures - even if that one step above is just to "document" what is out there (rather than doing additional measurements and analysis) - I value the workflow that does not utilize guesswork.

 

 

[Fegato]

I think "push a button" is a long way off. After all, we'd all want a different button! (take the same base data, and we all produce and want to produce something slightly different)

Anyway - BlurX does a marvellous job of leaving all the tiny galaxies alone, as does StarX. I'm in the middle of a hell of job trying to mask these all before some heavy stretching to avoid super bright / saturated galaxy blobs all over the place! I need GalaxyX please....

[saac]

24 minutes ago, Fegato said:

I think "push a button" is a long way off. After all, we'd all want a different button! (take the same base data, and we all produce and want to produce something slightly different)

Anyway - BlurX does a marvellous job of leaving all the tiny galaxies alone, as does StarX. I'm in the middle of a hell of job trying to mask these all before some heavy stretching to avoid super bright / saturated galaxy blobs all over the place! I need GalaxyX please....

I'm holding out for Cloud X

Jim

[ONIKKINEN]

I could see an automated preprocessing suite happening in the near future. Something that calibrates, stacks, gradient extracts, deconvolutes, colour calibrates and stretches an image automatically. All of those have an automatic tool available today, just not in the same package so not a stretch at all to believe something like this is around the corner.

I think only positives will come with a tool like that. The hobbyist that just wants to observe the final image with as few steps as possible will be pleased with the ease of use, and the enthusiast that wants to tweak every little thing can just not use it - or better yet, use the automation and spend more time tweaking and fine tuning other things in the image.

[tomato]

4 hours ago, vlaiv said:

For me, there is clear distinction between the two.

In case of stacking images - there is precisely defined mathematical workflow that has been proven to be correct.

In second case - it's much like giving someone a list of numbers and asking them to guess which number comes next. Sure, it is educated guess (AI has to be trained) - but it is still a guess and as such - not 100% right all the time.

When solving a mathematical equation, you don't say - here, this is solution, but I'm 87% confident it is the right one , If you apply correctly mathematical principles - you can be 100% sure you have the right solution.

Since I consider astronomical imaging to be more than producing pretty pictures - even if that one step above is just to "document" what is out there (rather than doing additional measurements and analysis) - I value the workflow that does not utilize guesswork.

 

 

But who processes data by a 100% precisely defined mathematical workflow? If we did wouldn’t we all get precisely the same result if we started with the same data? The irony is that if we want to process the data purely by a defined mathematical based workflow, wouldn’t this best be left to a machine?

[Stuart1971]

Here is what it does for me, my stars are not to bad, but I do have a tiny bit of elongation in two corners, I think it’s tilt, and it’s driven me mad for a while now, as  the more I try to adjust it, the worse it gets, and was starting to take the fun out of the hobby, as I could just not solve the issue, and my eyes were drawn to the poor corner stars every time I looked at my images,  I was fairly happy with BluX 1 and AI2, but this BlurX version 2 with AI4 is just incredible, here are the example images 

3 examples from 3 corners of the same image, the 4th corner was perfect anyway.

Raw is just stacked with no adjustment, then BlurX 1 with AI2 and finally the newly released BlurX 2 with AI4,

like I said they are not too bad to start with really

 

Edited December 15, 2023 by Stuart1971

[vlaiv]

12 minutes ago, tomato said:

But who processes data by a 100% precisely defined mathematical workflow? If we did wouldn’t we all get precisely the same result if we started with the same data? The irony is that if we want to process the data purely by a defined mathematical based workflow, wouldn’t this best be left to a machine?

Most professional astronomers do data reduction in such way.

If you go to any server and download data - you'll get nicely calibrated and reduced data with complete workflow of how it's done - so that you know what sort of data you are using.

So that is first stage - use well known and deterministic algorithms up to point where you have linear stack ready for processing.

Next stage is automated as well in some cases - we just don't think about it. Every time you take your phone or your digital camera and do following:

1. Set auto

2. Shoot image

3. Download jpeg and/ or send image to printer

That is what happens - predetermined, exact sequence of processing steps is taken to produce uniform looking image. Not only uniform looking - but also "correct" image. You will 100% agree that image that you are looking that is taken by digital camera is what you are seeing with your eyes.

[tomato]

I suppose the problem is we can’t use the own eyes test on most Astro based images.