Using AI

[Andrea Zullo]

I don't edit with Pixinsight because I just started and I'm still learning with PS. Everyone tells me that Pix is better, and I will probably use it in the future, but I would like to deepen my knowledge first. However, I would like to make a comment. I know many astrophotographers, more or less expert, some even of a certain importance, and they all agree that using blurx (and similar) means introducing details not present in your image. The problem is actually a little more extensive if AI is abused. Now, since this is an established fact, why don't we differentiate the two processing methods? Those who use AI to make photos prettier should be in a league of their own. I'm not saying stop progress, but I don't think it's fair to compare the two. It is not a comparison like the analogic vs digital photo was, but using AI really means skipping processing processes (for convenience?) and idealising your photos by adding details. I find it similar to when you use cheats for video games: at first it's cool, but then you lose interest in the game. I hope to spark a discussion battle 🙂 everyone (or almost everyone) knows it, but it's better to stay quiet because good photos get likes

[Astro Newbie]

Hi Andrea, 

You make a good point, however, I think this should only be a consideration when it comes to competitions when there could be an advantage using such tools.

Artefacts (artificial details) in images can also be introduced using software like PixInsight if you are not careful selecting the right parameters. I don't think it's always easy to know just how far you can go whether you are using AI or not. Again this only ought to be a consideration within a competition. I don't consider it a problem with images in the gallery as long as it is mentioned that use was made of AI tools.

Were it not for all the vast array of tools available to us such as Photoshop, our astrophotography images and any photos we took would most likely end up looking much the same. The really great images I see posted here are as a result of artistic talent and tools such as as Blurx only make a tiny contribution. 

Thanks for starting this discussion, looking forward to reading more comments. 

Cheers and CS, 

Ray 

 

[Andrea Zullo]

Hai ragione, però la creazione di artefatti accidentali nello sviluppo normale di una foto di solito non è un effetto voluto. Si cerca di evitarlo. Usando alcune IA invece si è ben consapevoli che la creazione finale è fittizia. Sono d'accordo con te che bisognerebbe indicarlo sempre. Quello che mi sorprende è l'omertà di tanti utilizzatori.

[Christopher Curtis]

Any photograph (and that includes astrophotos) is an interpretation. We make lots of choices of what images to capture, what to show and how to show it, what to edit out, what to emphasise and how to process the data we collected. What choices and processes are acceptable depends on why we are making a photograph and common expectations. Adam Block's brilliant talk digs into this here: https://telescope.live/tutorials/ancient-echoes-forgotten-sky

Personally, I don't think applying artificial intelligence in image processing is essentially any different to any other process we use. What matters is WHY and HOW those processes are used. Of course, BlurXterminator and similar AI processes can produce things that are not "real" in the data - but that's true of any process. For example, any stretched image has completely different pixel brightnesses, in a different relationship to each other, than was there in the data collected by the sensor. That's the whole point - we can't see anything, otherwise, but it's a subjective choice how we make the image appear.

Many people would say that astrophotographers should only apply processing to the data where mathematical and statistical algorithms are applied, not that it would be OK to directly edit individual pixels (painting in or out) but few would worry about someone doing exactly that to add a label or notation or if we know that it is intended to be a work of imagination. The key is the intention. Trying to fool people or misrepresent things is not usually acceptable but honestly attempting to interpret or express something is.

[Reggie Jones]

Very interesting topic for discussion.  I would generally agree with Christopher Curtis' comments - generally, I use all of PI's tools when processing data to try to tease out the detail that is already contained in the data.  There are many tools in PixInsight to do this and many of them are not AI developed tools such as Multiscale Median Transform or Multiscale Linear Transform.  Where I think you invite lots of trouble is trying to introduce information in your image that was not there originally.   

[Byron Miller]

I think the "AI" tools are a life saver for consume astrophotography including the images we get from telescope live.  I like the Takashi FSQ 106's but they're not edge to edge round stars on full frame and "AI" BlurXterminator in correct only mode will correct the aberrations in the stars and give me a flat field.  I've done my own math and ran my own observatories - there is probably $70-100k of gear and time on each of those scopes and even with all the passion and knowledge of people processing images - i don't think it's wrong to use a MODEL (not really AI but i guess it is) that is trained on aberration and can correct for aberrations we consumers face - such as non-flat fields, curvature, star elongation, back-spacing, sag/flexure, or simply that QHY600s are almost too big for a reduced FSQ106 and if you want to make a beautiful MOSAIC, being able to run BlurrXterminator to clean up the edges for a tighter alignment is a miracle.

I'd wager that blur exterminator based on modeling of optical issues is more "precise" than MMT transformation which leans in on artist preference in some ways 🙂

I've just learned I don't give a hoot. As long as this passion/hobby is fun then that's all that matter. We're using tech to collect the photons, tech to save the photons, tech to colorize and make things look pretty - makes sense to use tech based on modeling of data to correct aberrations in data.

PixInsigth developers would typically ban plugins/discussions there of if the workflow of these ML/AI tools added data but they love what RC-astro has done... it's like we're getting correctable optics and mirrors but in software 😉