Using StarNet++ with StarTools on linear data

[Heno]

I am a bit confused here. Maybe this tutorial needs an update to reflect 1.8 version.
1, 2, 3 and 4 is fine, no problem. (ST will load the starless image if it is just opened and then saved using Image Magick)
14. Reloaded the compose module, LRGB files, kept, told ST that the image was non linear to prevent tracking, binned, cropped to get the same size as the initial (starless) image.
15. Loaded Heal module, star mask, kept. (Image looked all black as it was not stretched yet, but the stars were gone).
Pressed Tracking to turn tracking back on.
Launched Autodev, then Wipe when I realized that the image was now monochrome.

15. Next time I tried I told ST that the image was linear. But how do you turn tracking off if you tell ST you have a linear image? I can't figure that one out.
Same process again. bin, crop to have same image size, launch heal, remove stars (I see them disappear), keep. Set the mask to full or none, same difference.
Launch Autodev. THE STARS ARE BACK.
If I load this as non-linear, there is no color, and I think I understand why.
But if I load it as linear, the stars are back when I open Autodev. Are the Heal module only working on the luminance channel or what.
Is this procedure only for OSC images?

I am probably stupid, but I just can't see what I'm doing wrong.

Helge

[Cheman]

I have found that StarXTerminator does a better job for me than does Starnet++. I assume the process would be the same, but if there are any differences or improvements to this in Startools, I would be interested in what Ivo has to say. It is not free, but I think it results in less artifacts left by the star removal. There is a free trial
Che

[Heno]

I have found that StarXTerminator does a better job for me than does Starnet++. I assume the process would be the same, but if there are any differences or improvements to this in Startools, I would be interested in what Ivo has to say. It is not free, but I think it results in less artifacts left by the star removal. There is a free trial

I will certainly give that a try, thanks, but unfortunately Starnet is not the main problem for me.

[admin]

Using any sort of algorithm based on neural hallucination will dream up different results for each L/R/G/B stack, and compositing four such individually treated stacks will yield color artefacts at the very least.

If you have LRGB data, the best you can is to use the procedure on just the luminance ("L") data (or a synthetic luminance composite if you so wish).

The procedure to incorporate StarNet++ in this manner, is predominantly meant for mask generation purposes, and is arguably a far more valid (documentary) use of the AI than dreaming up detail that does not exist (itself derived a training set of data you never actually acquired).

If you absolutely must remove stars, at least the Heal module can be made to guarantee any replacement pixels are as neutral as possible; the Heal module was specifically designed to minimise the influence the "made up" pixels have on neighbouring pixels once put through the various algorithms in StarTools (including deconvolution). Second, the Heal module can be made to guarantee replacement pixels are dimmer than the ones they replace. This means re-compositing of stars and starless image can be done perfectly with a "lighten" operation to arrive at the 100% original image.

This avoids the dreaded "translucent star" syndrome (where people erroneously use a throttled additive operation/layering).

Best is to ask yourself whether removing stars is the best course of action. The answer is almost always "no". Unless, of course, you wish to simply present a starless rendition.

[firebrand18]

I generally avoid separating stars and background as try and keep images as 'documentary' as possible, even when playing with messy bortle 8 data. I have played with Starnet ++ v1 to see how it fares (mixed results).

Curious if anyone has seen this post on AB this morning, as Starnet ++ v2 is out and apparently leaps ahead of v1 and better than StarExterminator.

https://www.astrobin.com/forum/c/astrop ... version-2/

[almcl]

I likewise don't often remove stars,l but had a play with the new V2 of Starnet.

I found that:

• It's about 10 times faster than the original
  
  does not require mucking about with .bat files (I hate command line interfaces and scripts and avoid software that requires them)
  
  produces better results than I got with V1, although some of this maybe down to switching to a Mak Newt as the original couldn't deal with Newtonian diffraction spikes properly or sometimes at all.
  
  Doesn't offer much in the way of user adjustable settings

Here's a quick working of the Rosette Nebula, (I was actually trying out a Hubble palette) see what you think?

AS as hubble starless .jpg (111.56 KiB) Viewed 18705 times

[Cheman]

Curious if anyone has seen this post on AB this morning, as Starnet ++ v2 is out and apparently leaps ahead of v1 and better than StarExterminator.

I've tried V2 and did get better results. Not any better than StarXTerminator which I give a slight edge to. But close, and still free. .

[firebrand18]

@almcl looks pretty good!

My main concern with this fiddling around is "breaking the chain" of ST Tracking Engine by flipping in and out with the data; I see the real value of ST by giving it pristine Linear data, un-touched except for stacking, and letting the "flow" work its magic until the end when tracking is turned off and you have a "final" result; after that, you can fiddle around all you want with post-ST processing to fine tune.

Many others outside of ST do lots of manipulations before inputting a "linear" (?) file into ST; don't think that is the way to go but maybe I'm off-base here.

Can't believe @Mike in Rancho hasn't weighed in yet...slacking off!

[almcl]

I actually ignored the 'only stretch the image' advice and fed the fully processed StarTools output to Starnet++V2 to produce that result. If I was going to drop the stars back in afterwards, it might be necessary to extract the stars early on and then process them separately.

[Mike in Rancho]

@almcl looks pretty good!

My main concern with this fiddling around is "breaking the chain" of ST Tracking Engine by flipping in and out with the data; I see the real value of ST by giving it pristine Linear data, un-touched except for stacking, and letting the "flow" work its magic until the end when tracking is turned off and you have a "final" result; after that, you can fiddle around all you want with post-ST processing to fine tune.

Many others outside of ST do lots of manipulations before inputting a "linear" (?) file into ST; don't think that is the way to go but maybe I'm off-base here.

Can't believe @Mike in Rancho hasn't weighed in yet...slacking off!

Moi? Oh dear.

I already feel bad sometimes when I check the first page for the most recent topics, and I see five "Mike in Rancho"s sitting there. Clearly I'm littering the forum with too many posts and questions. Ivo will never catch up.

I tried starless back a ways. Mostly because I guess it's all the rage, and still is, to split the image and process stars and other stuff independently. Still a bit dubious even if one were trying hard not to manipulate the data improperly. But I used both SN++ v1 and ST heal methods, and I didn't really come up with either an improved workflow or final image. Probably worse on both accounts.

I think it may be more useful for those who want the stars "out of the way" to then use a heavy hand with range masking and that soft megablurring, that sort of thing.

Personally I'm okay with some fully starless images, if one wants to show a reasonable facsimile of what a nebulous region looks like. Wide-field Polaris IFN is probably the most obvious example where this yields a very interesting rendition and maybe even helps the viewer "see" what is going on out there in extragalactic space.

I think in the past year I've only processed stars and image separately one time, that being a too-short-integration Ring Nebula where the target and the stars+background could not take the same stretch. Or more accurately the stretch I wanted to apply. It was a hack job though.

And now we get SN++ v2, which can apparently destar a linear file. But what do you then do with it? Even beyond the fact that the stars may well have lost their proper diffraction effects now, can SVD generate an apod star mask, when there are no stars...