RC BlurXterminator …. Any comments ?

[Startrek]

I’m a bonafide ST user however I recently saw a video on the latest RC plug in for PI , BlurXterminator
On face value it looks very impressive but time will tell , like all plug ins
The procedures to use and apply it effectively puts all the responsibilities back on to the end user ( typical PI )
Im interested if anyone has given the trial a go and how does it compare to the current SV Decon in V1.8

Comments most welcome

Martin

[admin]

I saw some results a few days back. It doesn't appear to be bonafide deconvolution (and if that assessment is indeed true, it would bother me quite a bit that it is billed as such), but rather very similar to Topaz Sharpen AI, however trained on astrophotographical material.

One giveaway that this is not appear to be true deconvolution, but rather the same sort of "re-interpretative" affair that we have seen is in Topaz AI, is that point lights are not truly refocused; stars don't become brighter. They just appear to change shape. On astrobin I have also seen a "before" and "after" image where stars and detail appeared out of thin air (though the problem is that you would really want to be able to compare linear before and after images to ascertain any foul play).

Because it has been trained on astrophotographical material, the "deep dream" artefacts that it still appears to generate are more plausible, however appear still very much present (Adam Block's video on YouTube showed a comparison between a Hubble image and an "RC Enhanced" image and the detail could not be fully corroborated - he, however, did not seem to think it was a big deal which is strongly felt was disappointing).

I kind of don't want to support this sort of thing, but you can find the video at https://youtu.be/_rWEuW4MTrQ?t=3357

In that same video, the BXT seems to be "cleaning up artefacts" in drizzled images, which is definitely not the purview of a deconvolution algorithm and would be consistent with my earlier assessment that its main job is to dream up signal - plausible as it may be - that was never recorded.

Another big turn off / red flag for me, is that it is apparently needed for the AI to distinguish between "stellar" and "non-stellar" detail (there is no such subjective interpretative step necessary for real deconvolution and it is unclear why/how this is needed or performed).

We seem to - once again - be dealing with a parameterless tool that augments an image with "signal" that was never recorded, as opposed to putting you in control of a (reversable!) mathematical transformation that is based on statistical and physical modelling of the same data.

Don't get me wrong - like Topaz AI, it clearly performs deconvolution-esque corrections, however these corrections seem to happen across a continuum of real, plausible and fantastical with no clear way of knowing when (or how/why) this happens. "Trust me, it's AI" is not cutting it and neither does "trust me, this time it was trained on astronomical data".

I guess we are firmly in the era where people think it necessary to invent and mould their own reality, rather than record what's in front of them. It makes me (and it seems Juan / PI on the PI forums) quite sad. It's the exact same as teenage girls using AI Instagram filters to look like flawless models, and it fundamentally preys on the same insecurities. Let your data's beauty speak for itself. It doesn't need this sort of thing. I can only imagine how Juan must feel all this happening on his own platform.

That said, in the event that I am totally off base with my analysis, I too would be quite interested seeing a comparison between ST's real (Richardson & Lucy - based) deconvolution and - what appears to be - "Instagram" deconvolution.

[Mike in Rancho]

Uh oh [goes and turns off my instagram filters...]

j/k, don't have an instagram. Not a teenage girl neither.

Yes, the latest PI rage! I've sort of had the feeling sometimes they are trying to follow ST's lead, at least for written plugins. GHS? And now this perhaps-deconvolution, that does claim to take into account spatial PSF variance, as well as spatial error or distortion.

The RC webpage for BXT seems like he claims there's data fidelity in what he is doing - possibly trying to distance himself from the Topaz-like issues? I wouldn't know how to otherwise evaluate.

A couple concerns I had was the way he discussed PSF, and that there was no need for iterations. Unsure if that meant truly no iterations are run, or the module determines how many are needed and just does it with no user intervention.

I have three CN friends I regularly PM who are trying BXT out right now. So far I've just seen a couple crops of zoomed side-by-sides, one as a before/after and the other I believe compared PI's own deconvolution (i.e. don't use this) to BXT. There was improvement no doubt. I thought it worked better on the smallest stars, though all were resolved inward some even the big ones. I'll have to look again - yes I didn't check to see if there was brightening along with the "shrinking." I also haven't yet seen a result on non-stellar detail.

We have not yet swapped any of these files for comparison versus ST. And yes...I mentioned that we should do so.

PI being PI and deconvolution, as well as BXT, being a linear operation, I'm not sure if they have a screen stretch preview to see what they are doing? Regardless, after running BXT, they will still be using PI for their global stretch - maybe with masks or possibly GHS - and thus they are likely to have what we would probably call FilmDev stars for the bigger ones.

[Mike in Rancho]

Oh I did want to toss in a comment a bit on AI as well...

There was a big to-do on CN lately for that also. But, I'm not sure there's necessarily a problem with AI in and of itself, but rather what it actually does that's important.

So, an AI computer could be made to play chess, and to always keep learning. In fact I think they've done that! But of course it's constrained by the rules of the game.

I could see the same in AP, though it may take some fun out of the whole thing. Thus, Ivo could perhaps write an AI that mines the stacked dataset, and then goes and pre-fills all the ST module options and settings to what it thinks, based on astro training, is likely to produce the "best" final result. And you could either accept or reject those suggestions. Or maybe just push one button and watch it go (like STReplay except not a replay). That would still be playing within the rules of the game, using ST's existing modules.

But if the AI creates a new rook out of thin air and drops it on the chessboard, well then that's painting/clone stamping/cheating. Same in AP.

I dunno if I'm thinking about it right. Seem reasonable?

[admin]

The RC webpage for BXT seems like he claims there's data fidelity in what he is doing - possibly trying to distance himself from the Topaz-like issues? I wouldn't know how to otherwise evaluate.

A couple concerns I had was the way he discussed PSF, and that there was no need for iterations. Unsure if that meant truly no iterations are run, or the module determines how many are needed and just does it with no user intervention.

It reads to me like it's a typical one-pass impulse->response consultation of a neural net, without any actual traditional deconvolution step happening. E.g. it is not the case of determining the PSF and helping a traditional decon algorithm to reverse-apply this PSF while keeping noise propagation in check.

It may be entirely by accident, but the description on BXT's webpage to me definitely reads like a "this-is-totally-not-Topaz-AI" piece.
Particularly the - paraphrasing according to my impression - "deconvolution is an ill-posed problem" therefore "anything goes" allusion is worrying (and would be quite wrong if meant that way). The "it "understands" what astronomical structures actually look like at finer scales than can be resolved using amateur equipment" is setting off big bat-symbol-in-the-sky alarm bells for me. There is nothing special or different about structures that cannot be resolved by amateur equipment. It would be a monumental astronomical discovery if it were! There is no such magical threshold beyond which things need to be treated differently.

The Centaurus A image on that page definitely displays the "this-is-what-you-want-to-see, right?" magically thinning structures (which cannot be corroborated in closer up images.) so reminiscent of the Topaz "dog hair" filter. The choice of a galaxy to demonstrate the module's prowess is somewhat unfortunate, as they are objects where detail is much easier to resolve further in higher quality images (rather than, say a nebula in our own galaxy that becomes more diffuse at larger scales).

E.g. what should be structures that "fall apart" into smaller structures of diffuse dust when better resolved, with BXT become solid/defined thinning interconnected strands.

The non-resolving of the stars (which by and large remain fuzzy) vs other parts of the image that become better resolved than said stars (which are supposed to be the smallest resolvable detail) is also quite jarring/atypical for deconvolution.

I have three CN friends I regularly PM who are trying BXT out right now. So far I've just seen a couple crops of zoomed side-by-sides, one as a before/after and the other I believe compared PI's own deconvolution (i.e. don't use this) to BXT. There was improvement no doubt. I thought it worked better on the smallest stars, though all were resolved inward some even the big ones. I'll have to look again - yes I didn't check to see if there was brightening along with the "shrinking." I also haven't yet seen a result on non-stellar detail.

We have not yet swapped any of these files for comparison versus ST. And yes...I mentioned that we should do so.

That'd be interesting to see.

PI being PI and deconvolution, as well as BXT, being a linear operation, I'm not sure if they have a screen stretch preview to see what they are doing? Regardless, after running BXT, they will still be using PI for their global stretch - maybe with masks or possibly GHS - and thus they are likely to have what we would probably call FilmDev stars for the bigger ones.

There is indeed a non-permanent "screen stretch" you can apply on top while you're processing in the linear domain. For comparison purposes, you can keep this stretch constant and also apply it permanently if you wish.

But, I'm not sure there's necessarily a problem with AI in and of itself, but rather what it actually does that's important.

That's entirely my stance as well. I think AI is amazing and it will always be my first love (I studied it for a while in University), but I don't think people understand what is happening when they use it.

In AP, when AI is used in this unsophisticated way, you are no longer the sole contributor of the signal when you use it. It's that simple. Whether that's a problem (and to whom) is a different story. It's ethically no different to airbrushing. Airbrushing is not a problem in itself. But it is a problem if you are trying to deceive people.

Assuming you are engaging in documentary photography, I don't regard your image well if I detect its use. At the very least, I will think less of your skills as an astrophotographer, because you have proven to be a poor custodian of your signal. I may even think less of you as a person if I know that you actually fully understand what you did to your signal and did it anyway and still claim it as a documentary photograph. Now you're actively trying to deceive me and trying to make yourself and your skills look better than they are.

Different would be if an AI would establish the right PSF(s) to reverse-apply. No new signal or information would be added and you can fully articulate why you transformed your signal in that way and how (e.g. using a generally accepted and mathematically reversible algorithm). The only thing the AI would do in that case is act as tool to model the distortion. It would not engage in any interpretative black box signal modification.

[Startrek]

Thanks for honest and forthright comments

A number of folk on our local forum don’t care how or why but as long as it “seems” to work , then it must be useful and good

Narrow minded ? Who knows , everyone chooses their own pathway

My summary….
Mathematics vs hypotheticals
Science vs Art
Fiction vs Non Fiction
Commonsense vs Nonsense


We do live in a world where folk get bored with fundamentals and search for something that’s not real or realistic

I’m so thankful for people like Ivo who keeps Startools real and honest
A true and bonafide image processing application which keeps getting better and better

Thanks
Clear Skies
Martin

[firebrand18]

This is a thread touching on my personal perspective of capturing and presenting astro photos to people, many who have no idea what is involved, what is real and what is not. I only use ST for astro processing and getting deeper and deeper into understanding how it works, using advanced functions and loving the results of how it keeps things "real" and "documentary" if you stay within the rules and don't start wandering off in unorthodox ways.

Fully agree with Ivo's sentiments here and find it very confusing (irritating?) browsing Astrobin and seeing images with similar equipment/sky conditions as mine but with wildly different results in detail, color, etc., most using PI / PS. SHO presented as such with only Ha/OIII?

From the BlurX AB link I read and Ivo shared here, notice the very short but to the point comment from (respected) Andrea Tasselli:

"Let's cut the chase and make everything artificial and AI generated. At least we'd save on the glass expenses..."

Anybody else feel this is bang-on and unfortunately where things are headed? I tried Topaz DeNoise a while back out of curiosity and after the initial "wow, looks great" moment, once you start looking closer, makes a mess of stars and have not used it since.

BTW, ST 1.9 Alpha is a winner IMHO. Processed 12 hrs of M31 from Bortle 8 with UV-IR and the results are striking compared to earlier versions, with new Compose Filter options, Wipe Correlation Filter improvement and SVDecon updates awesome. I find star sampling 'starfishies" @Mike in Rancho is actually better in SVD than 1.8; even though it seems to "miss" many seemingly good candidates as many have mentioned in recent posts, however, as Ivo states, starts to make sense for reasons apparent when you look at the end result.

Can't wait for 1.9 main release!

CS

Nick

[dx_ron]

I've been following the (spirited) threads on BX and looked at what side-by-side 'with vs without' comparisons I could find. I am hampered by not being familiar with PI (not to mention being fundamentally ignorant about the inner workings of image processing as a whole).

What I see typically reminds me of what I see with the Highlights/Shadows Detail Boost effects of HDR. The difference is that I can give an "explain it to a non-astrophotographer"-level explanation of what ST is doing (please correct me if this is wrong!): ST is changing the stretch levels/curves/gamma parameter/whatever-is-the-under-the-hood-reality at very localized scales in the image to increase very local contrast.

BX, on the other hand, is using an AI engine to do "stuff", which seems to result in similar localized sharpening.

Did PI not have any such highly local contrast/sharpening tool? Is that part of the explanation for why BX is going to take over the world?

[Startrek]

BXT is a hot topic at the moment on our local Astro forum , many comments about this new tool , folk posting images, mostly positive feedback.( Most members use PI, APP and PS etc… )
I try to push the ST barrow when and where ever I can , even to the extent of showing my before and after images using V1.8 SV Decon.I don’t receive too many comments about SV Decon , folk are simply caught up in this BXT tool due to a whirlwind of media exposure.
Maybe ST should have more exposure to the media to show just how damn good this unique type of image processing really is.
Waiting in anticipation for V1.9 in early 2023
Clear Skies
Martin

[admin]

For those interested, I was made aware of some comments by RC on the PI forums regarding the training process;

It confirms my observations and fears that it's all sadly very black box and that there is no actual real AI-assisted deconvolution going on, but that it is merely a basic reinterpretative "Topaz AI Sharpen for astrophotography" affair. There is indeed a lot of conjecture - even on behalf of RC - as to what the neural network is actually doing;

" More training of this AI2 beta version has probably increased the deconvolution accuracy over AI1. The results posted so far seem to be consistent with this theory."

and

"The network is probably trying to correct for uneven FWHM in the color channels."

Such is the nature of this type of simple re-interpretative/hallucinatory use of a neural net, that the end result remains subjective and unquantifiable in terms of its actual workings and thus the validity of the "improvements" contained within the end result.

It's a real shame (and frankly rather grating) that RC keeps calling it "deconvolution" while it is ostensibly not provable that it is so, and does not appear to behave like real deconvolution would behave.

To turn this into real AI-assisted deconvolution, RC could possibly try recover the "PSF" that the AI used in the frequency domain (given that real deconvolution is simply division in the frequency domain) and then feed that PSF to a real deconvolution routine. However, I have a feeling that that the recovered PSF will need a lot of cleaning up, if at all usable (because I have a feeling it won't look like a typical astronomical PSF at all...)