Can masked stretch be done in ST?

[BainthaBrakk]

Hi Ivo,

First: I processed(a bit fast maybe ) the only data I had on the subject to try to help with the problem OP had. That was the reason, certainly not to complain about Startools.

Regarding autodev, it might be that I just haven't understood how to use it, but looking at the processed pictures above by you I doubt it. *I* don't like the star profile that autodev creates. That doesn't mean it's not correct in any way.

You ask why I remove stars? Easy: In some areas of the sky (star rich areas) I wish the nebulosity to play the "lead role". Therefore I process the stars less aggressively by themselves. By doing so I get tighter stars and the nebulosity gets to shine. I know that this is not your philosophy, but there are a lot of "more seasoned" astrophotographers that think the same way as I do on this subject.

I also know that there are a lot of ways to reduce stars in Startools, but I have found that I get results that *I* like better by using other software when doing this.

When it comes to "neural hallucinations" (funny btw) I mostly agree with you even though it is a different subject entirely. As I had limited data I certainly pushed a bit too hard, but this was not to present my best picture in any way, it was just to prove another point. Even then, I don't think it has introduced any crazy structures that it certainly can do when used too aggressively.

Regards -

/Ulf

[Mike in Rancho]

Well, to hop back in...

It really becomes un-debatable when you start framing things in that way. I could say that *I* like using Affinity photo to paint little smiley faces in my images. Well, it may be very bad astrophotography practice, but you can't debate what someone claims they *like.* And it's just that a lot of these practices really are more "art based on astro pictures, some more loosely than others" than AP. If that's your thing, no one can stop you. And as is often the case, many might find that art to be quite beautiful. But it should be flagged as such, and the procedures shouldn't be shared around unless it is clear that it is indeed, art.

But really my only beef with your initial post was that you shoehorned, slightly disparagingly IMHO, properly ethical AP techniques as "absolute scientific accuracy." I don't buy that, and it's really just a way to open the door to wholesale manipulation and give it cover. Well, since I'm not a scientist, I might as well turn this whole area blue! Come on.

You are right that a lot of seasoned imagers/processers have jumped on the art train. Starnet is now built into PI, and starless processing (by necessity filling in holes with fake detail) has become a thing for almost any image, instead of a "break glass in case of emergency" situation where you could argue it needs to be used. Likewise, whether via well-known YT astro personalities, or experienced APers, selective range and color manipulation is also a popular thing, particularly in PI. The fact that many are doing it, or are asking how to do it (you should see the current target discussion on my other forum), does not elevate to anything more than artsy selective manipulation. I find it disappointing that the "PI look" has become mainstreamed.

But if one wants to do art, to each their own. I just fear that many beginners are picking up these techniques from others and not realizing that they are just making stuff up.

That said, not having any trained eye Spidey senses, I am also having trouble seeing the problem between the APOD close-up and the comparative image that, assumably, has been warped. Of course I have no idea how the APOD was created and processed either. It is lacking in green for Ha, if a SHO, though that could be mapping. But the other version seems mostly matching as to areas of detail and color - it just mostly seems that one was done with a top notch telescope and camera, and the other was done by a backyard amateur, with limited equipment, maybe not the best tracking or guiding, and probably OSC files that went though Bayer interpretation and had to be binned down.

Ivo, what should the trained eye be seeing there?

[BainthaBrakk]

Hi Mike,

But really my only beef with your initial post was that you shoehorned, slightly disparagingly IMHO, properly ethical AP techniques as "absolute scientific accuracy." I don't buy that, and it's really just a way to open the door to wholesale manipulation and give it cover. Well, since I'm not a scientist, I might as well turn this whole area blue! Come on.

I have no other agenda than helping the OP, I really don't know what you are talking about, sorry.

I find it disappointing that the "PI look" has become mainstreamed.

I wouldn't know. I use Startools and don't even own PI.

But the other version seems mostly matching as to areas of detail and color - it just mostly seems that one was done with a top notch telescope and camera, and the other was done by a backyard amateur, with limited equipment, maybe not the best tracking or guiding, and probably OSC files that went though Bayer interpretation and had to be binned down.

This is downright condescending and if you are trying to help Ivo here, you are doing an awful job.

Regards -

/Ulf

[admin]

Hi Ivo,

First: I processed(a bit fast maybe ) the only data I had on the subject to try to help with the problem OP had. That was the reason, certainly not to complain about Startools.

Not taken as such at all!

Regarding autodev, it might be that I just haven't understood how to use it, but looking at the processed pictures above by you I doubt it. *I* don't like the star profile that autodev creates. That doesn't mean it's not correct in any way.

AutoDev creates a starting point for you to build on as you please. It retains as much true information as possible and does not unceremoniously white clip it. This is particularly important when applying deconvolution (in StarTools that is), which is the #1 way to coalesce stellar profiles into point lights. From there (e.g. after decon), other techniques can follow.

You ask why I remove stars? Easy: In some areas of the sky (star rich areas) I wish the nebulosity to play the "lead role". Therefore I process the stars less aggressively by themselves. By doing so I get tighter stars and the nebulosity gets to shine. I know that this is not your philosophy, but there are a lot of "more seasoned" astrophotographers that think the same way as I do on this subject.

To be honest, I think most astrophotographers have that same goal, but there is no need to resort to making up nebulosity to accomplish this.

When it comes to "neural hallucinations" (funny btw) I mostly agree with you even though it is a different subject entirely. As I had limited data I certainly pushed a bit too hard, but this was not to present my best picture in any way, it was just to prove another point. Even then, I don't think it has introduced any crazy structures that it certainly can do when used too aggressively.

Believe it or not, "neural hallucination" is actually a good way to describe the technique that is used by the Topaz AI suite and Starnet (for examples of the term is used, see here and here).

To force the neural net to come up with plausible guesses, the network is asked to hallucinate based on peripheral cues and a measure of injected noise.

I can see a fair bit of hallucinated detail in your image that is plausible, but does not actually exist in the "real" image by Martin.

[Mike in Rancho]

Hi Mike,

I have no other agenda than helping the OP, I really don't know what you are talking about, sorry.

No worries, but not much I can do then, it was your own quote.

This is downright condescending and if you are trying to help Ivo here, you are doing an awful job.

Was only trying to follow the comparative samples provided, and here, it is I who don't know what you are talking about.

It is my understanding that the first-referenced image, the APOD, is in fact the second, lower image from which a crop had been taken. At least that appears to be the case if I open the APOD link in another window and then hit +.

I have no idea where the second dataset, which I presume is the first-shown crop image, came from, but is apparently the one using techniques that change the integrity of the data. Nor do I know what software Ivo ran on it in order to create those data errors for this comparison. It is only said that it is 4.5 hours and I believe your post was 3 hours, so it seems to have come from somewhere else. Ivo's personal data? Did not mean to condescend that!

My point was that I can't see the wrongly-introduced details. To repeat myself, to my eye everything seems to match up reasonably well as to location and color, and the only conclusion I would draw is that it was taken with a much more normal level of equipment than the APOD was. Are those Planewave's not off-the-charts? I have to also figure they have mounts to match.

So, where is the disconnect? It seems we are not understanding each other here at all.

[admin]

@Mike in Rancho
The two images are a comparison of an excerpt of Ufl's image (3 x 1.5 = 4.5h) and the corresponding area in Martin's 30h image.

There are a lot of discrepancies.

When I look at an image, then what goes through my mind - granted, I'm not a typical viewer - is the following;

As said, I get suspicious right off the bat when stars are deformed in an atypical way. Stars are bloated and "swallow" each other.

Also puzzling, is that, what remains of the destroyed / white-clipped stellar profiles, suggests acquisition through a cheap(er) DSLR lens that was stopped down slightly; there is a slight 6-spike pattern visible, typical of such a configuration (see here for an intentional Bokeh example with a higher quality lens). Cheaper lenses tend to have a smaller amount of blades making up the aperture. Having a cheap DSLR lens in the optical train when the dataset is a SHO dataset is... suspicious. This is rather circumstantial, and there may be other explanations for this strange diffraction pattern. However, it often points to AI-induced detail generation (with an AI trained on lots of DSLR data). Particularly if there seems to have been separate deliberate attempt to obfuscate the stellar profiles, as is the case here.

Green arrow; Looking at how fine the detail is, do the stars match? They do not. At all. A non-overexposing star is per definition the smallest detail that was resolved in your dataset. Yet the stars are resolved far worse than the too-good-to-be-true fine detail. Some stars are so smudged that they barely even show up, yet there is tons of dark "detail" many times finer that somehow registered against a darker background where the bright stars did not.

Red arrow; Are there gross discrepancies in detail resolved? And/or do areas that should be high in signal show markedly worse resolved detail than areas of high signal. Yes and yes.

With one or more red flags like these, it's usually time to find a reputable comparison source.

Yellow arrow; (and probably the most obvious to a layperson). Has detail been altered wholesale? Sure has. Detail has clearly been replaced with other plausible (neurally hallucinated) detail. Some has been removed, some has been added. The structures pointed out look nothing like the more canonical 30h dataset.

Errors.png (239.67 KiB) Viewed 5395 times

To some, all this may seem inconsequential. To others these manipulations are rather bewildering.

One thing is for sure - one is space photography, the other is space fantasy (and even then, the fantasising was left to an AI!).

We live in a strange time where some think image processing is preparing a dataset so an AI can better hallucinate. And even going as far as recommending to avoid a tool (such as AutoDev) because the AI cannot replace that tool's true recovered detail as effectively otherwise.

I for one sure liked things better back in 2014...

[Mike in Rancho]

Thanks Ivo,

That is exactly what I was looking for, and it is amazing what your trained eye can pick up. Whereas to me, it seemed just not a semi-pro Planewave, leaving me perplexed. I am thinking it could be rather difficult for beginners or even intermediates to recognize such discrepancies.

That said, I apologize to you and Ulf for any obnoxiousness, misunderstanding (apparently I can't do 3 x 1.5 = 4.5!), or unintended insults. I will give myself a time out.

[opestovsky]

I processed(a bit fast maybe ) the only data I had on the subject to try to help with the problem OP had.

Thanks.
How fast was "fast"?

If it isn't too long, could you do my data, in the first post?

[BainthaBrakk]

I processed(a bit fast maybe ) the only data I had on the subject to try to help with the problem OP had.

Thanks.
How fast was "fast"?

If it isn't too long, could you do my data, in the first post?

Hi!

I am sorry, I am less than useless at processing HOO-palettes from OSC-data. What I was suggesting was that you remove the stars from your "quick and dirty"-version with Starnet++ and then adding stars back from processing the stars separately. That is a bit hard for me to help with as I don't have the cropped dimensions.

Regards -

/Ulf

[BainthaBrakk]

As said, I get suspicious right off the bat when stars are deformed in an atypical way. Stars are bloated and "swallow" each other.

Well a reason for that might definitely be the difference in resolution between my equipment and a Planewave 17.

Also puzzling, is that, what remains of the destroyed / white-clipped stellar profiles, suggests acquisition through a cheap(er) DSLR lens that was stopped down slightly; there is a slight 6-spike pattern visible, typical of such a configuration (see here for an intentional Bokeh example with a higher quality lens). Cheaper lenses tend to have a smaller amount of blades making up the aperture. Having a cheap DSLR lens in the optical train when the dataset is a SHO dataset is... suspicious. This is rather circumstantial, and there may be other explanations for this strange diffraction pattern. However, it often points to AI-induced detail generation (with an AI trained on lots of DSLR data). Particularly if there seems to have been separate deliberate attempt to obfuscate the stellar profiles, as is the case here.

Ivo, I think this is a good example where you are trying to find evidence of tampering when there really is none.

The "6-spike" pattern is a result from pinched optics of my Esprit 100. These appear sometimes in triplets/quadruplet scopes when photographing in sub-zero temperatures. You might not be familiar with that phenomenon in Australia I guess.

We live in a strange time where some think image processing is preparing a dataset so an AI can better hallucinate. And even going as far as recommending to avoid a tool (such as AutoDev) because the AI cannot replace that tool's true recovered detail as effectively otherwise.

I mostly use Autodev. When I get a problem, I try something else. Sometimes (mostly with mono data) it is really hard to get a result I find appealing when using it. That doesn't mean I think it is bad. It might be that I am doing something wrong, I have some stacked files you could have a look at that really shows this.

Regards -

/Ulf