Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

PixInsight StarNet Process for SCTs

Astrophotography Imaging SCT DSO
  • Please log in to reply
9 replies to this topic

#1 astrodj111

astrodj111

    Lift Off

  • *****
  • topic starter
  • Posts: 2
  • Joined: 23 Sep 2017

Posted 18 September 2021 - 07:36 AM

I see all these great images where the stars have been removed using StarNet within PixInsight. Unfortunately, it doesn't work for me. I'm using a Meade LX200 12" either at F/10 or F/6.3.  I believe my stars are too large for starnet to handle it.  However, I think I found a work around (I documented it here in case you have similar challenges https://www.chaoticnebula.com/pixinsight-star-removal-with-starnet/) .  

 

My original image

image-17.png

 

After using StarNet

image-18.png

 

If I use Integer Resample to downsample the image , StarNet works great. Of course I lose all of the detail I get with my imaging rig. So I do the following

  • downsample my image 4x.
  • Use StarNet to create a star mask.
  • Upsample the mask 4x.
  • Create a range mask from my upsampled star mask to smooth everything out.
  • Use the range mask as my star mask and do morphological transformation and convolution

Results!!!

image-41.png

 

 


  • rayp likes this

#2 WadeH237

WadeH237

    Cosmos

  • *****
  • Posts: 7,707
  • Joined: 24 Feb 2007
  • Loc: Ellensburg, WA

Posted 18 September 2021 - 08:27 AM

Nice!

 

I usually image at smaller image scales and have always been unimpressed with Starnet, due to the artifacts that it leaves behind.  Partway through summer, I switched to a wide field setup and was quite surprised to see the Starnet was much, much better at the wider image scale.

 

I am looking forward to trying your technique when I get a chance.



#3 H-Alfa

H-Alfa

    Viking 1

  • -----
  • Posts: 827
  • Joined: 21 Sep 2006
  • Loc: Spain

Posted 18 September 2021 - 09:49 AM

I see your images have an STF application. Is it because they are linear? If yes, you must first to make them nonlinear. An standard STF to HT usually works well. For difficult cases I would suggest you to stretch a little less.
  • pfile and hornjs like this

#4 pfile

pfile

    Fly Me to the Moon

  • -----
  • Posts: 5,750
  • Joined: 14 Jun 2009

Posted 18 September 2021 - 10:47 AM

re the STF - that is a really good point, i missed this when OP posted this in the PI forum. of course it can still be stretched and then have an STF applied.

 

as for SN performance, my understanding is that Nikita trained SN on ~500mm refractor images so it's not a huge surprise that it works better on images with similar image scale...

 

rob



#5 astroian

astroian

    Ranger 4

  • -----
  • Posts: 393
  • Joined: 21 Sep 2016
  • Loc: Swindon, UK

Posted 18 September 2021 - 12:08 PM

I’ve not had any problems using Starnet++ with my Edge HD 11 images. As H-Alfa says you need to make the image non linear first.

Cheers,
Ian
  • hornjs likes this

#6 Jim Thommes

Jim Thommes

    Fly Me to the Moon

  • *****
  • Posts: 6,300
  • Joined: 20 Sep 2004
  • Loc: San DiegoCA USA

Posted 18 September 2021 - 04:57 PM

I have not had trouble with starnet leaving star residuals except for stars that are bright enough to have a halo. Then I see the halo after starnet has removed the stars.

 

Having said that, I like your downsample technique. Rather than using a Morphological transformation, I would probably use a defect map process. To use defect map effectively, the starmask should be full intensity or zero for each pixel - no smoothing.



#7 Alex McConahay

Alex McConahay

    Voyager 1

  • *****
  • Posts: 10,632
  • Joined: 11 Aug 2008
  • Loc: Moreno Valley, CA

Posted 18 September 2021 - 09:42 PM

I have lots of residuals with my 12 inch RC......the vanes just kill it, I think. Or maybe I am doing it all wrong. 

 

Alex


Edited by Alex McConahay, 18 September 2021 - 09:43 PM.

  • SiriusFritz likes this

#8 astrodj111

astrodj111

    Lift Off

  • *****
  • topic starter
  • Posts: 2
  • Joined: 23 Sep 2017

Posted 19 September 2021 - 06:27 AM

Tried using StarNet with linear and nonlinear images. Results are always the same.  Based on my understanding of StarNet is that the neural network it uses was based off of refractors and doesn't work well with reflectors. 

 

@Jim - Definitely going to see what DefectMaps will do for star removal. Looks like an interesting option. 



#9 Linwood

Linwood

    Mercury-Atlas

  • *****
  • Posts: 2,587
  • Joined: 09 Jul 2020
  • Loc: Ft Myers, Florida, USA

Posted 19 September 2021 - 09:29 AM

I've had similar issues to the OP, and also image a bit over sampled (0.288"/px).  I've had decent luck removing and then adding back the same stars, but awful luck removing narrow band stars and adding back LRGB, the artifacts around them are bad.

 

Now that said, on my list is to stop trying to do it post RGB or LRGB combine, instead doing it on the individual channels (this presents the issue that I usually RGB combine linear for PCC, but I found a pixelmath script that simulates an unstretch). 

 

But yes... star bloat just spoils starnet for me.  Star mask actually works better (with a bit of MT adn convolve thrown in). 

 

Are there weights files for different purposes?



#10 H-Alfa

H-Alfa

    Viking 1

  • -----
  • Posts: 827
  • Joined: 21 Sep 2006
  • Loc: Spain

Posted 19 September 2021 - 10:10 AM

Tried using StarNet with linear and nonlinear images. Results are always the same. Based on my understanding of StarNet is that the neural network it uses was based off of refractors and doesn't work well with reflectors.

@Jim - Definitely going to see what DefectMaps will do for star removal. Looks like an interesting option.

Can you share your original MasterLight in xsif format? I would like to try to apply starnet to it to see what the issue is.

Enviado desde mi VOG-L29 mediante Tapatalk


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics





Also tagged with one or more of these keywords: Astrophotography, Imaging, SCT, DSO



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics