Starnet++ v2 is here!

For those who know what Starnet++ is, this is great news, you can get it here:  Instructions can be found in the included README file.

For those who are unfamiliar, read on….

What is starnet++??

Starnet++ is a neural network tool that can remove stars from the background of your images.  It can be used as a standalone tool or as a module within PixInsight. Since it’s initial release, Starnet++ has become a invaluable tool in many people’s processing workflows.  It can be used to remove the stars from your image after stretching the image so that the subject of the image can be enhanced without blowing out the stars. Finally, the stars can be added back into the final stages of the workflow.    

Here’s an example of what Starnet can do- 

Rosette starless comparison


Issues with V1

  • Artifacts – Starnet, does an incredibly good job of removing stars however in many cases it does tend to leave behind blotchy artifacts or rings around where some larger stars were prior to removal.  
  • Performance- Starnet can take foreverrrrrrrrrrrrr to run unless you a) have a very powerful CPU and lots of memory, or b) configure Starnet to use your computer’s GPU.   In my case, making this change took execution time from about 10 minutes to 30 seconds.   
  • Only works on non-linear (stretched) data.  

Improvements in V2 

  • There are far less artifacts in the starless images produced by V2.  There is a new feature that will upscale your image prior to star removal & then scale the starless version back down.  This appears to produce a better result but does take significantly longer to process. 
  • There is a new feature that will allow starnet to process linear data, this might be very helpful for creating masks to be used for initial stretch operations.  
  • Performance can still be improved with the GPU modification, this is something that PI plans to make a native feature in a future PI release.  A really good walkthrough on how to do this can be found here-

On the left, a close-up of starnet v1.  On the right, starnet v2.  There are far less artifacts.  

2 thoughts on “Starnet++ v2

Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes:

<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>