PixInsight Starnet artefacts

[oymd]

Is there a way to avoid these sorts of artefacts in Starnet in PI?

I've noticed its always happening with the brighter bigger stars, like here in IC1318, SADR, and also in IC6960 in the Western Veil?

In PI, Starnet has no settings to change or fine tune?

What could possibly be done to avoid these sorts of artefacts?

[vlaiv]

I don't think it is possible.

With very bright stars - there is often "halo" or "skirt" associated. These look too much like background nebulosity and I don't think that software "knows" how to remove them.

You could possibly try to teach it, although I'm not completely sure I can explain how to go about it.

StarNet++ uses neural network in its core - meaning you can teach it how to remove stars. In order to do it - you need to train it - this means that you need to give it bunch of image pairs - one with stars and one without stars.

See the problem - if we had starless version to begin with - we would not need StarNet++ .

Kidding aside - you would need to know how to synthesize halo/skirt feature on bright stars. I think easiest way to train StarNet++ is to take high resolution images with very tight stars (like Hubble images) and run median filter on them for star removal and run blur on them to get fatter stars as one is expected to get with amateur equipment.

Another approach would be to generate some sort of fractal type nebulosity images and then add in random star fields on top of that.

[oymd]

14 minutes ago, vlaiv said:

I don't think it is possible.

With very bright stars - there is often "halo" or "skirt" associated. These look too much like background nebulosity and I don't think that software "knows" how to remove them.

You could possibly try to teach it, although I'm not completely sure I can explain how to go about it.

StarNet++ uses neural network in its core - meaning you can teach it how to remove stars. In order to do it - you need to train it - this means that you need to give it bunch of image pairs - one with stars and one without stars.

See the problem - if we had starless version to begin with - we would not need StarNet++ .

Kidding aside - you would need to know how to synthesize halo/skirt feature on bright stars. I think easiest way to train StarNet++ is to take high resolution images with very tight stars (like Hubble images) and run median filter on them for star removal and run blur on them to get fatter stars as one is expected to get with amateur equipment.

Another approach would be to generate some sort of fractal type nebulosity images and then add in random star fields on top of that.

Vlaiv, in PixInsight, Starnet has no options or settings to change, train or do anything with it. 
 

It’s just an icon process that you apply to your starry image. 

[vlaiv]

2 minutes ago, oymd said:

Vlaiv, in PixInsight, Starnet has no options or settings to change, train or do anything with it. 
 

It’s just an icon process that you apply to your starry image. 

Yes, but that is because Starnet is not part of PI - it is stand alone open source software.

You can read more about it here (github repository for it):

https://github.com/nekitmm/starnet

[Lee_P]

53 minutes ago, oymd said:

Is there a way to avoid these sorts of artefacts in Starnet in PI?

I've noticed its always happening with the brighter bigger stars, like here in IC1318, SADR, and also in IC6960 in the Western Veil?

In PI, Starnet has no settings to change or fine tune?

What could possibly be done to avoid these sorts of artefacts?

I get these artefacts too, and clone them out using Photoshop.

[Richard_]

I'm not sure if it will solve your problem, but there is another way to create star masks in Pixinsight. Check out the EZ Processing Suite provided by Darkarchon:

https://pixinsight.com/forum/index.php?threads/ez-processing-suite.14937/

There is an "EZ star reduction" script  which has a pretty cool star mask creation tool. It can automatically create a star mask (a la Starnet++) which you can further refine to suit your needs. This includes convoluting/deconvolution, adding ellipses to cover halos surrounding stars etc. 

You can use that tool to generate star mask(s) without actually performing the star reduction process (just cancel out after creating masks). As an FYI, I believe that this tool only works with non-linear data, so you'll have to stretch your image before running the script. 

Give it a go and let us know if it works for you. 

[ollypenrice]

It might be time to try Russ Croman's StarXterminator which many say beats Starnet. It's on my list of jobs to do.

Olly

[Sp@ce_d]

2 hours ago, ollypenrice said:

It might be time to try Russ Croman's StarXterminator which many say beats Starnet. It's on my list of jobs to do.

Olly

Oh I just noticed there’s now a version for Pixinsight too.. excellent. Time to try it out.

[oymd]

11 hours ago, Richard_ said:

I'm not sure if it will solve your problem, but there is another way to create star masks in Pixinsight. Check out the EZ Processing Suite provided by Darkarchon:

https://pixinsight.com/forum/index.php?threads/ez-processing-suite.14937/

There is an "EZ star reduction" script  which has a pretty cool star mask creation tool. It can automatically create a star mask (a la Starnet++) which you can further refine to suit your needs. This includes convoluting/deconvolution, adding ellipses to cover halos surrounding stars etc. 

You can use that tool to generate star mask(s) without actually performing the star reduction process (just cancel out after creating masks). As an FYI, I believe that this tool only works with non-linear data, so you'll have to stretch your image before running the script. 

Give it a go and let us know if it works for you. 

 

3 hours ago, ollypenrice said:

It might be time to try Russ Croman's StarXterminator which many say beats Starnet. It's on my list of jobs to do.

Olly

 

46 minutes ago, Sp@ce_d said:

Oh I just noticed there’s now a version for Pixinsight too.. excellent. Time to try it out.

Yes, I’ve just watched Visible Dark’s YouTube video where he compares StarXterminator with Starnet in PixInsight. 
 

It seems Xterm does a much better job of removing stars with minimal halos, but he noted that the star quality might be a touch better with Starnet. 
 

Also, I was amazed at how FAST Xterm was compared to Starnet. 
 

My PC has a Ryzen 3950x with 16 cores, 64GB ram and an RTX 2070 super Nvidia GPU, and Starnet takes a long time. He did mention there is a hack to speed up Starnet for Nvidia cards, but not sure what it was. 

[david_taurus83]

StarXterminator is good but on bright stars with large halos, instead of leaving that hatched artifact Starnet does, SX blurs the background where the halo was so it looks a bit out of place compared to the background. Still needs some work.

[oymd]

Well, today I went to RC Astrophotography website to have a look and download the plugin.

$59 for a plugin within PixInsight??!!!

Whilst I recognise that a lot of work went into this plugin, and I appreciate the effort, I think that $59 is a LOT to ask for a plugin??!!