StarTools vs Siril - A challenge

[decay]

Hi Alex,

The issue has to be to do with the way that OptiDev stretches the brightness around those mid-to-bright stars.

That may be true. Have you tried playing with the brightness tranche bit-depth, as Ivo suggested? Sometimes it shows a noticeable impact, at least for the highlights.

Another idea would be to to try FilmDev in order to find out whether it is really OptiDev which is causing this.

Best regards, Dietmar.

[ajh499]

Hi Alex,

The issue has to be to do with the way that OptiDev stretches the brightness around those mid-to-bright stars.

That may be true. Have you tried playing with the brightness tranche bit-depth, as Ivo suggested? Sometimes it shows a noticeable impact, at least for the highlights.

Another idea would be to to try FilmDev in order to find out whether it is really OptiDev which is causing this.

Best regards, Dietmar.

Hi Dietmar

Yes, I have tried playing with the bit-depth, it doesn't really seem to make much difference.
It initially dims the image, and possibly tightens the stellar core, but it doesn't alter the size of the halo once the image has been stretched to the same point as before.

I had a quick look at FilmDev but didn't finish the process as it didn't seem much better.
I'll have to take another look at try to quantify any differences

Alex

[ajh499]

OK, quick comparison

I have two images here
https://www.dropbox.com/scl/fi/tiqji1bl ... pxxo4&dl=0

and here
https://www.dropbox.com/scl/fi/1xohcrbz ... xgtuj&dl=0

They have just been stretched and coloured (since the "haze" seems worse once the colour is included), and no decon. The OptiDev is just default setting, no ROI, but 16-bit brightness. The FilmDev, I tried to match the brightness of the brighter parts of the nebula, but it's just an estimate.

If you toggle between them in an image viewer, the OptiDev one feels to me much more hazy, I think due to the lower contrast and the star halos. The FilmDev one obviously has large blobby stars and has lost detail in the shadows.

If you zoom in and toggle between image looking at the stars, the OptiDev halo, is about the same size as the blob of the FilmDev version (no surprise there I guess, StarTools is doing what it should), but I find the the large "flat" halo looks unnatural and weird, like the stars have just been pasted on top of the background noise.

Maybe it's just me

[hixx]

Hi Alex
no, I don't think it's You, but the data. OptiDev treats brightness levels equally throughot the image, it does not differantiate between a "star signal" or a "nebular signal" Consequently, it the image looks "as-is", without any human interpratation. There is no intelligence telling OptiDev to "recognize" a star as point source, this is where SVDecon would come into play.

[hixx]

I found, when developing an image, there is no silver bullet tool that magically brings up a Hubble-like image. Instead, for me it worked best not to use just few tools to extremes. (I am sure everyone here might have different personal approaches which work best for them)
After Wipe, I just use a decent Optidev stretch with ROI to work out the object (DSO) then a tiny bit of Contrast to pull down darks a bit, HDR for local stretch enhancements, SVDecon just a few iterations and after Colour a bit of Superstructure (Isolate or DimSmall) to push back stars & maybe NBAccent. Rarely i use Flux or Layer which are really vast tools.... Playing around is the fun part, trying differnt settings for each tool and how they work on the object. There are so many ways to achieve different goals in ST, after some time You'll develop a feeling what works best for You and how to reach an intended goal. Of course, AI could do the same in a second, but that's just without the fun part!

[decay]

Hi Alex,

Well, some time after posting, I already had the feeling, that this was not my best idea … sorry.

Afterwards, I had a go (or two) with your data and I can confirm that the bit-depth setting is of no use in this case.

And yes, creating adequate and comparable OptiDev and FilmDev versions is nearly impossible. But instead blinking your versions is a great example of what OptiDev is able to achieve. For me, the stars are quite nice under control and I’m afraid, the remaining halos are just an unavoidable result of data quality (like discussed before).

I know, we are not able to influence weather and seeing conditions and we have to take what we get. And we want to get the most out of it. But in order to do a fair ‘challenge’ is it not right to compare StarTools results with AI based processing results. StarNet cuts out the stars and replaces them by synthesized background data. Instead, we would have to compare the result we get from StarTools to any other software/processing which respects signal integrity. Maybe Siril processing without using StarNet?

I agree with you and Ron, that StarTools’ stars often show a very distinct look, different to what we see in most other cases. I’m not sure what it is, maybe it’s worth to have a closer look (like Ron did).

I like your image you posted at the end of page 1 of this thread and the ‘OptiDev’ result of the comparison you did in your last post. If you would manage to get the best of both versions together, this would be a respectable result considering this limited data quality.

And: Maybe there is no reason to choose between StarTools and some other software. If you got high(er) quality datasets, you can process them with StarTools and be proud of your result which was processed with respect to signal integrity and which represents – well, reality and truth. And if there’s only very limited data and it took month to get even that – why not cheating if it’s OK for you? It’s a hobby. The only thing is that other people may be fooled when publishing such images without hint of what was done.

Not sure, if this helps at all …

Best regards!
Dietmar.

[ajh499]

There is no intelligence telling OptiDev to "recognize" a star as point source, this is where SVDecon would come into play.

I think I often find that although SVDecon does an amazing job (better than any other software that I've tried), it just accentuates the difference between the now even more point-like star in the centre, and the big coloured hazy blur around it.

I found, when developing an image, there is no silver bullet tool that magically brings up a Hubble-like image. Instead, for me it worked best not to use just few tools to extremes. (I am sure everyone here might have different personal approaches which work best for them)
After Wipe, I just use a decent Optidev stretch with ROI to work out the object (DSO) then a tiny bit of Contrast to pull down darks a bit, HDR for local stretch enhancements, SVDecon just a few iterations and after Colour a bit of Superstructure (Isolate or DimSmall) to push back stars & maybe NBAccent. Rarely i use Flux or Layer which are really vast tools.... Playing around is the fun part, trying differnt settings for each tool and how they work on the object. There are so many ways to achieve different goals in ST, after some time You'll develop a feeling what works best for You and how to reach an intended goal. Of course, AI could do the same in a second, but that's just without the fun part!

That's my problem, I spend hours fiddling with StarTools images and rarely see anything other than their flaws - stars especially. Maybe this data isn't good enough, but I have better data, and I'm not sure that I'm that much happier with how they come out.

I'm not sure more data would make a difference either - I've tried increasing rejection from this stack, too see if tighter, rounder stars help, but it doesn't make much difference to the stars, either better from the tighter focus, or worse by reducing the total exposure.

Instead, we would have to compare the result we get from StarTools to any other software/processing which respects signal integrity. Maybe Siril processing without using StarNet?

Yes, maybe. I don't know how to process the stars and background separately in Siril without using StarNet, and with them combined the stars will probably just bloat the same as they do with FilmDev or "traditional" software.
I don't think StarNet is doing quite what we think though, at least one of the features that Ivo identified is actually in the original stack prior to any processing, not sure why StarTools missed it. If I have a chance, I'll look through all the processing steps that I took and see if I can see where artifacts were introduced.

I like your image you posted at the end of page 1 of this thread and the ‘OptiDev’ result of the comparison you did in your last post. If you would manage to get the best of both versions together, this would be a respectable result considering this limited data quality.

Thanks, that's encouraging There's actually not much difference between those images, just more processing steps involved in the first one - and it used Ron's suggestion of undoing decon on the brigher stars.
Doesn't the plain OptiDev version look flat and blurry compared to the FilmDev one though? It certainly does to me when blinking between them.

Alex

[hixx]

Hi Alex,

think I often find that although SVDecon does an amazing job (better than any other software that I've tried), it just accentuates the difference between the now even more point-like star in the centre, and the big coloured hazy blur around it.

That's why I don't let Decon go all the way, I tend to use a less aggressive setting with fewer iterations. Additionaly I use Shrink modules with Color Taming and few iterations. There is a problem with stars embedded in nebulas: due to the extreme contrast, the edge will always look a bit "processed", so I tend to less agressive settings.
In terms of aquiring signals, I work with a DSLR and use subs between 20 to 60 s, but hundreds of those. That definately helps with signal quality. The steps for me were:

[hixx]

1) Identify the optimal amp setting of Your camera. (some modern "dual gain" sensors have an analog Amp before the A(D section which helps gain without sacrificing Dynamic or Noise). You can try finding more sensor detail on DXOmark.com For My Sony A7 III it's ISO 800, there's an (old) table in the manual for some common DSLRs.
2) Identify the optimal sub exposure time (which makes just overexposes few stars and gives a "30% histgram peak", as Ivo described.
3) Sort out bad subs and stack at least a total integration time of 4-6 hours
I found 300 -400 subs deliver great results on brighter, common DSOs, You'll need more sub exposure time /number of subs for dimmer objects

[ajh499]

Hixx

Sorry for the delay in replying, I've been away from home.

I've found the same as you, I need to back off the SVDecon a bit otherwise it gets a bit too agressive.

As I mentioned earlier in this thread, I've used spreadsheets to estimate the correct sub-exposure and gain for my ASI183MM Pro. 30 secs to 60 secs seems to be about right for Lum, and two or three times that for RGB.

I maganed to get a bit more data for the Rosette, but had to ditch a load of it as the auto-focus didn't run properly while capturing, so there's only a little more exposure time, but I'm a bit more happy with the following

LRGBHa3-forum.jpg (265.71 KiB) Viewed 493 times