Beware of star fixing tools

[tomato]

I imaged the Veil Nebula last night with my SY135/RC571c/NBZ combination and with my usual philosophy of "get an image rather than spend the clear sky time tinkering with the kit",  I put up with some terrible star shapes on the LHS of the image. No problem, I thought, BlurXterminator and Startools Repair module will sort them out, and to be fair they did a reasonable job. However when zooming in to look at how well they had worked their magic I discovered they had done this to Epsilon Cygni:

This is what it looked like on the original stacked image:

The software has managed to convert the flares from the brightest star in the image into a very convincing PN! It is strange that the software has done a good job in tidying up all of the other stars in the image but this one was just a step too far...

 

[windjammer]

I think you will have to paint the blur out and mask in some part of the original!

BTW is your collimation/focus dialled in - some asymmetrical flaring on the star images of the original stack ?

Simon

[ollypenrice]

'Twas ever thus.  Take the bulk fix and do the rest by hand.

Olly

Edited July 28, 2023 by ollypenrice
False click.

[Adam J]

My view with this sort of extream of AI processing is that if you are willing to do this, you may as well be willing to use a tool that takes the contense of a star catalogue and artificially generates the star feild from the data.

My honest opinion is that your end result simply looks fake. It chatches the eye as odd in the same way as early CGI, too perfect too symetric.

Adam

Edited July 28, 2023 by Adam J

[tomato]

52 minutes ago, Adam J said:

My view with this sort of extream of AI processing is that if you are willing to do this, you may as well be willing to use a tool that takes the contense of a star catalogue and artificially generates the star feild from the data.

My honest opinion is that your end result simply looks fake. It chatches the eye as odd in the same way as early CGI, too perfect too symetric.

Adam

That is certainly one point of view. Yes it is fake in the sense that the original stars were much more numerous and suffered from optical artefacts.  The stars were reduced and reshaped to (in my opinion) create a more pleasing image of the Veil Nebula than if I had left them untouched. It raises the question of how far do you go? Is it OK to correct eggy or slightly trailed stars, but if the software can fix more serious defects do you use it or consign the data to the bin and wait another 3 weeks for a clear night?

If I lived in Atacama I'm sure I would rely more on the raw data and would certainly have the clear sky time to fettle the optics to wring out their best performance. Until then I will continue to use all of the processing tools available to me, but hopefully to create a result that still owes more to the raw data than AI, and we all know how fast that technology is developing. 

I took Olly's advice and "fixed" the offending Epsilon Cygni, now posted in the Deep Sky imaging section.

 

2 hours ago, windjammer said:

I think you will have to paint the blur out and mask in some part of the original!

BTW is your collimation/focus dialled in - some asymmetrical flaring on the star images of the original stack ?

Simon

Yes,  the collimation (sensor tilt) is off, and maybe the focus, hence the reliance on the software fixes.

 

 

[Adam J]

1 hour ago, tomato said:

The stars were reduced and reshaped to (in my opinion) create a more pleasing image of the Veil Nebula than if I had left them untouched.

The thing is that, it's very difficult to know what the AI is really doing. In the end it's goal is to show you what it thinks you want to see. But I think it's a mistake to think that it is "reshaping" the stars. It's much more likely that it is simply sampling what is present and replacing it (not reshaping it) with a 2d gaussian profile. In that sense it's closer to being computer generated than it is enhanced. But that for me is not the issue, the issue is that it looks computer generated. 

So what would I do with that data. I would prefer a starless image to one with fake looking stars added back in to it. 

It's like my wife using AI filters on her photos to change how she looks. Yeah she looks younger, but a younger fake looking version of my wife. Hence, I tell her I would rather look at the real thing than something I know is not real. 

Just my view. 

 

Adam 

[windjammer]

I had a look at star tools web site - I hadn't heard of it till now (duh) - from the description given, AI isn't mentioned.  It sounds more like a deconvolution tool.  If so, what would be your objection to that ?

[Tomatobro]

Those collimation issues down to the ring spacer on the mirror?

[tomato]

I used BlurXterminator, the AI based tool first, principally to sharpen the nebula, on the default settings. The stars on the left hand side had pronounced wings which BXT dutifully rounded and tightened them so each star took the appearance of a triple star, rather like a Micky Mouse hat. I used the StarTools repair tool, option redistribute, average brightness for core location, and the three stars became one. So in my book, the processing brought the stars closer back to their ‘real’ appearance. 
You could argue that I shouldn’t be imaging with such a poorly optically optimised set up, but I did so for the reasons explained previously. I don’t care for 100% starless images, I would rather have them present, enhanced and manipulated, than leave them out altogether.

[tomato]

4 minutes ago, Tomatobro said:

Those collimation issues down to the ring spacer on the mirror?

No, this is the Samyang 135 lens, I think the issue is due to camera sag, my agricultural camera support being the culprit. I feel another little engineering project coming on…☺️

[ollypenrice]

10 hours ago, Adam J said:

My view with this sort of extream of AI processing is that if you are willing to do this, you may as well be willing to use a tool that takes the contense of a star catalogue and artificially generates the star feild from the data.

My honest opinion is that your end result simply looks fake. It chatches the eye as odd in the same way as early CGI, too perfect too symetric.

Adam

This opinion has been advanced in one way or another since Blur XT appeared. I think it may have its origins in the fact that Russ Croman trained his software on Hubble data, leading some people to conclude that it refers to Hubble images when processing the user's image. It doesn't.  Bear in mind that Hubble has imaged a vanishingly small part of the sky over its entire lifetime so transcribing Hubble images into amateur ones wouldn't work at all. The Hubble images to do this don't exist. Croman used Hubble data because it is so good that any training performed on it will be as accurate as possible.  Blur XT refers only to the image on which it is being applied. It performs an analysis of that image and then makes adjustments based on that analysis. These adjustments can be called guesses but they are very, very highly educated guesses.

I find Blur XT simply gives tighter stars and a touch of sharpening. The image remains wholeheartedly the image.

Olly

[ollypenrice]

8 minutes ago, tomato said:

No, this is the Samyang 135 lens, I think the issue is due to camera sag, my agricultural camera support being the culprit. I feel another little engineering project coming on…☺️

I think you should certainly make two of these things, dear boy!

lly

 

[tomato]

16 minutes ago, ollypenrice said:

I think you should certainly make two of these things, dear boy!

lly

 

Will do, it will be a lathe and milling machine job rather than a 3D printer, still Old School on some things.😉

[ollypenrice]

Just now, tomato said:

Will do, it will be a lathe and milling machine job rather than a 3D printer, still Old School on some things.😉

I was only joking! As for 3D printing, I hate, loathe and despise it!!!

Olly

[tomato]

1 minute ago, ollypenrice said:

I was only joking! As for 3D printing, I hate, loathe and despise it!!!

Olly

I keep looking at 3D printers, but with molten plastic vs machined alloy the latter always wins in my head. It must be great though, seeing something being printed on the bed when it all works, a bit like AP.

[ollypenrice]

2 minutes ago, tomato said:

I keep looking at 3D printers, but with molten plastic vs machined alloy the latter always wins in my head. It must be great though, seeing something being printed on the bed when it all works, a bit like AP.

You can drill plastic with metal. You can't drill metal with plastic...

Olly

[Elp]

You can also 3D print in metal. And building materials, and even food. I'm sure a drill will go through all of it though.

[Adam J]

2 hours ago, ollypenrice said:

This opinion has been advanced in one way or another since Blur XT appeared. I think it may have its origins in the fact that Russ Croman trained his software on Hubble data, leading some people to conclude that it refers to Hubble images when processing the user's image. It doesn't.  Bear in mind that Hubble has imaged a vanishingly small part of the sky over its entire lifetime so transcribing Hubble images into amateur ones wouldn't work at all. The Hubble images to do this don't exist. Croman used Hubble data because it is so good that any training performed on it will be as accurate as possible.  Blur XT refers only to the image on which it is being applied. It performs an analysis of that image and then makes adjustments based on that analysis. These adjustments can be called guesses but they are very, very highly educated guesses.

I find Blur XT simply gives tighter stars and a touch of sharpening. The image remains wholeheartedly the image.

Olly

No I know its not taking the stars from the Hubble images that would be ludicrous and thats not how it works. But the issue with AI is that its not really possible to know what its doing. I am surprised that he used hubble images through as that would have in my view taught the AI to add diffration spikes to the stars.

What I am saying is that it has created a mathamatical model of what a star should look like based on what its been shown (unsuprisingly a gausian profile) then using that model it is replacing the stars with a AI generated stars that conform to that model. Hence it is not adjusting the stars or fixing them. It is sampling them to determin what the star colour and the luminocity profile should be and then mathematically generating a replacement.  There is no need for it to refare to another image because its been taught what a star should look like.

If it was looking at the distortion you see in the orignal image, modeling it and then adjusting it would never result in the bright star being removed as it would apply the same model to that star.

It probably uses a totally seperate model (not because it was told to but because it has learnt to) to operate on nebulocity or glaxies etc. What it has dont with that star is fail to identify it as a star remove it for some reason and then treat it with the nebula mode. I cant say for sure of course but that is my hypothasis. The fact that it can turn a star into a planetry nebula however shows that a model is performing extrapolation and that Blur XT is not just purly acting on the data it has it is applying a model and generating new data that it is placing ito the image. Its done that because the AI made a mistake, but it is revealing none the less.

The problem with your hypothasis of how it works is that I have seen examples of bright nebulocity been turned into stars by Blur XT. I have seen a section of the witches broom recently posted on here were a fillement of nebula has been turned into a something that looks like stars on a string like a pearl necklace. Those stars dont exist, there is nothing for it to corrected, its added them from scratch into the image.

Ill go find you a good example of that.

None of that makes it invalid as a tool, even if I dont personally like the results, but I think its important to be aware that your resulting image is left with some arfificially generated elements within it.

Adam

 

 

Edited July 29, 2023 by Adam J

[ollypenrice]

1 hour ago, Adam J said:

No I know its not taking the stars from the Hubble images that would be ludicrous and thats not how it works. But the issue with AI is that its not really possible to know what its doing. I am surprised that he used hubble images through as that would have in my view taught the AI to add diffration spikes to the stars.

What I am saying is that it has created a mathamatical model of what a star should look like based on what its been shown (unsuprisingly a gausian profile) then using that model it is replacing the stars with a AI generated stars that conform to that model. Hence it is not adjusting the stars or fixing them. It is sampling them to determin what the star colour and the luminocity profile should be and then mathematically generating a replacement.  There is no need for it to refare to another image because its been taught what a star should look like.

If it was looking at the distortion you see in the orignal image, modeling it and then adjusting it would never result in the bright star being removed as it would apply the same model to that star.

It probably uses a totally seperate model (not because it was told to but because it has learnt to) to operate on nebulocity or glaxies etc. What it has dont with that star is fail to identify it as a star remove it for some reason and then treat it with the nebula mode. I cant say for sure of course but that is my hypothasis. The fact that it can turn a star into a planetry nebula however shows that a model is performing extrapolation and that Blur XT is not just purly acting on the data it has it is applying a model and generating new data that it is placing ito the image. Its done that because the AI made a mistake, but it is revealing none the less.

The problem with your hypothasis of how it works is that I have seen examples of bright nebulocity been turned into stars by Blur XT. I have seen a section of the witches broom recently posted on here were a fillement of nebula has been turned into a something that looks like stars on a string like a pearl necklace. Those stars dont exist, there is nothing for it to corrected, its added them from scratch into the image.

Ill go find you a good example of that.

None of that makes it invalid as a tool, even if I dont personally like the results, but I think its important to be aware that your resulting image is left with some arfificially generated elements within it.

Adam

 

 

The judicious imager needs to look at the results of BlurXT intervention very carefully, for sure.

Olly

[tomato]

49 minutes ago, ollypenrice said:

The judicious imager needs to look at the results of BlurXT intervention very carefully, for sure.

Olly

Absolutely, that was the reason for starting this thread in the first place.

[Adam J]

4 hours ago, ollypenrice said:

The judicious imager needs to look at the results of BlurXT intervention very carefully, for sure.

Olly

and star exterminator too. I have seen it remove small planetary nebula before and glowing edges to nebulosity too. For example I imaged the question mark nebula and by blinking it in PS between the before and after images I picked up and a shocking number of nebula elements that it had removed along with the star field. Looking in other people's images recently of that target there is a pronounced epidemic of missing objects within their images and it's all caused by blind trust of the star removal program. 

Adam

Edited July 29, 2023 by Adam J

[Elp]

Stitching software is also in the same boat, I tried MS ICE which has been lauded in a lot of places for mosaics, when compared to my manually aligned image it removed whole degrees of sky and put galaxies in wrong places.

[windjammer]

@ Adam

>> The problem with your hypothasis of how it works is that I have seen examples of bright nebulocity been turned into stars by Blur XT. I have seen a section of the witches broom recently posted on here were a fillement of nebula has been turned into a something that looks like stars on a string like a pearl necklace. Those stars dont exist, there is nothing for it to corrected, its added them from scratch into the image.

Ill go find you a good example of that.

<<

Go on then!

Simon

[Scott Badger]

I don't know of any processing tool that gives you perfect results, and no one, especially the developer, is claiming that BlurX does. And like any software in continual development, there may be bugs that need fixing, and according to a CL thread, it appears the latest releases of BX does have a problem, especially when used in the auto psf mode. When not broken and used properly, though, it can do deconvolution (the algorithms it's actually using) better than we can mostly because it can work with millions of parameters, tens of millions even, where we can only handle a few. But in the end it's still up to the operator to look at the result of any process, assess it, and decide whether it's an improvement. You can blindly trust the result of any processing software. Or not.

It *is* true that the nature of a trained neural network tool means a final result can't be 'predicted', or even back-engineered i.e. no matter how well you know the code and the training, you can't know exactly what leads/led to a particular result. But for our work, I don't see why that matters. I couldn't follow the math, let alone the code, in most of the processes we use, 'AI' or not. For research purposes, that opaqueness *is* a problem, but again, Russ Croman is the first to say that BlurX or any AI software shouldn't be used for that purpose. That said, professional astronomers have recently gained a better understanding of the M87 black hole by using AI to sharpen the image.

There is also the DIY vs DFY (done for you) divide, but for me it's like woodworking, there are some things I like to do with hand-tools, and others where power tools and guides are my choice. Positioning this as all or nothing, if you use BX you might as well just go out and get a Stellina and be done with it, is a straw man argument. It would be just as fair for an astrophotographer that uses film and manual guiding to level the same charge at anyone using electronics......

Arguments regarding image manipulation in general get muddy really fast, and the use of AI has little to do with it. Generally speaking, there's corrective and aesthetic; sharpening stars and fixing shapes is corrective, reducing the number of stars is aesthetic. That's pretty clear. But what about background and the use of tools like DBE and noise reduction, where's the line between correction and aesthetic improvement there? Or how we stretch images and insert contrast in a non-linear way, especially with tools like GHS? And then there's color...... I really don't see how any line drawn is anything but personal, or why anyone should feel compelled to process their images in a particular way. You can like, or not like, my images for all sorts of reasons, and the world goes on.

Cheers,
Scott

Edited July 30, 2023 by Scott Badger

[Adam J]

On 29/07/2023 at 20:42, windjammer said:

@ Adam

>> The problem with your hypothasis of how it works is that I have seen examples of bright nebulocity been turned into stars by Blur XT. I have seen a section of the witches broom recently posted on here were a fillement of nebula has been turned into a something that looks like stars on a string like a pearl necklace. Those stars dont exist, there is nothing for it to corrected, its added them from scratch into the image.

Ill go find you a good example of that.

<<

Go on then!

Simon

 

On 29/07/2023 at 20:42, windjammer said:

@ Adam

>> The problem with your hypothasis of how it works is that I have seen examples of bright nebulocity been turned into stars by Blur XT. I have seen a section of the witches broom recently posted on here were a fillement of nebula has been turned into a something that looks like stars on a string like a pearl necklace. Those stars dont exist, there is nothing for it to corrected, its added them from scratch into the image.

Ill go find you a good example of that.

<<

Go on then!

Simon

My issue is I have seen them in images from a couple of well known and liked forum users. I had cut some examples out of those images and was about to post then thought twice. As I don't want them to see it as a commentary on their image, they are still good images. In the end if you want to go looking take a look at recent Cygnus loop images and if you are sufficiently familiar with the target you will spot anomalies. 

I have decided that for me winning a technical debate is not worth it if it involves making someone I like feel bad about their image. I am willing to comment on not liking the result here as the OP opened a debate on the subject. 

One thing I might do is reproduce it myself but I am a busy man so that will be on my own time not just because you demand it of me. Suppose I could post it just to you via PM, bit I get the sense that in itself won't get me too far...