StarNet - new module for star removal

gvanhau

Well-known member
Jun 21, 2010
345
0
66
Santiago, Chile
Hello Nikita,

I've been testigin this module with excelent results.
I tested it on many different images with dissimile focal lengts and resolution including wide field images and it worked perfectly on all of them.

Thank you for making this available to us.

regards
Geert
 

astroimager69

Member
Mar 14, 2018
5
0
Hi everyone, i have just started using Starnet but keep getting black spots where the stars were (as in the attached image of the Heart nebula) . Any suggestions would be greatly appreciated
 

Attachments

SgtBilko

Member
Aug 27, 2019
8
2
Astrovetteman said:
I finally got Starnet++ to work in Pixinsight on my iMac running Catalina...only took about 6 hours of manipulating files and reading Forum posts!!  Am I correct that Starnet will ONLY work on 16bit unassigned TIFF files?
Thanks!
Tom
Tom

I am just about to try on a mac running Catalina.... to save me "...6 hours..." any chance you could list how you got it to work ?

Thank you
Ian
 

pfile

PTeam Member
Nov 23, 2009
5,709
116
SgtBilko said:
Astrovetteman said:
I finally got Starnet++ to work in Pixinsight on my iMac running Catalina...only took about 6 hours of manipulating files and reading Forum posts!!  Am I correct that Starnet will ONLY work on 16bit unassigned TIFF files?
Thanks!
Tom
Tom

I am just about to try on a mac running Catalina.... to save me "...6 hours..." any chance you could list how you got it to work ?

Thank you
Ian
i wrote the mac instructions in the readme file so you can ask here. but i hope i explained everything therein.

rob
 

mmisiura1

PTeam Member
Dec 14, 2017
27
1
astroimager69 said:
Hi everyone, i have just started using Starnet but keep getting black spots where the stars were (as in the attached image of the Heart nebula) . Any suggestions would be greatly appreciated
It would be useful to see original image as well, but I think that your image was excessively processed before applying starnet. You should apply it right after initial stretch ideally and process starless image separately.
 

msmythers

PTeam Member
Jan 16, 2013
1,179
1
www.astrobin.com
Hi Nikita...and Everyone ;)

I'm sort of back at least starting to do some processing again. New parts in the old computer so I can use Starnet++ the way it was meant to be used. I had been using the emulation trick a little bit but that got old and then so did I,haha.

Anyway here is a wide field shot of Cygnus. With the emulation method at a Stride of 160 this image took well over 2 hours to process. Now just 38 seconds. I've got a lot of catching up to do. Along with the wonderful new and updated tools in PI my old data is going to be getting a new lease on life.



Mike
 

Attachments

msmythers

PTeam Member
Jan 16, 2013
1,179
1
www.astrobin.com
Looks great Jim. I just finished adding the last piece of hardware into the computer rebuild/update. A small 256GB NVMe. This rebuild cost $380(motherboard/cpu/ram/nvme) and couldn't be more pleased. I didn't even have to reinstall Windows 10. Replaced all the hardware at once and booted. All drivers were added automatically and the Windows license did not have to be reactivated. I was shocked. The NVMe drive is being used just for temp stuff and the Windows page file. I'm fine with the OS on one of the other SSD's. I'm not overclocking this system either, just stock.

Anyway now I just have to get my head back in the game and start processing. Lots to relearn.


CPU Identification
CPU vendor ............. AuthenticAMD
CPU model .............. AMD Ryzen 5 2600 Six-Core Processor

System Information
Platform ............... Windows
Operating system ....... Microsoft Windows 10 Home
Core version ........... PixInsight Core 1.8.8-4 (x64)
Logical processors ..... 12
Total memory size ...... 31.950 GiB

Execution Times
Total time ............. 00:42.02
CPU time ............... 00:35.45
Swap time .............. 00:06.55
Swap transfer rate ..... 2531.147 MiB/s

Performance Indices
Total performance ...... 11194
CPU performance ........ 10678
Swap performance ....... 14019
 

vpcirc

Active member
Jul 9, 2012
34
0
pfile said:
the tensorflow dylibs need to be copied into either

  /usr/local/lib or /Applications/PixInsight/PixInsight.app/Contents/Frameworks/


since /usr is hidden in the finder, it might be easier to copy the tensorflow dylibs to /Applications/PixInsight/PixInsight.app/Contents/Frameworks/

rob

No such folder in windows 7. It seems to me there needs to be a DLL files, but there is none in the PI_OSX download. Also I cannot find any dylibs files on my computer. I actually had this working (can't remember how) then updated PI release and it disappeared. You are one of the smartest guys I know with PI, I still use your process icons on some things, but I'm totally lost (and apparently computer illiterate) I couldn't understand 1/2 the readme file. I need a "install starnet for dummys" set of instructions. I tried installing all the folders in the PixInsight bin folder (In a starnet folder and standalone) but it doesn't recognize the module. JP Metsavaino talked about this process at AIC and I think its outstanding. Using a starless image to apply structure in PS is amazing, 
 

pfile

PTeam Member
Nov 23, 2009
5,709
116
yes you are apparently using the wrong zipfile. OSX = mac OS X, not windows.

note that the install location of the weights files are slightly different on windows than on OSX. if PI starts up with the wrong "working directory" then you need to go into the pixinsight console and cd to the directory where the weights files are sitting, or else the module will not be able to find them.

rob
 
Jan 28, 2015
19
0
Astrovetteman said:
pfile said:
the command line versions only work on 16-bit tiff files. however, the pixinsight module can run on any view that you have open on the desktop, no matter the bit depth.

although starnet only works on stretched images, astroedo figured out how to run it on linear images by using STF to figure out an MTF curve to temporarily stretch the image in a way that can be undone.

rob
Thanks for the info....I just was corresponding with another member and he said the same thing.  I'm just going to play around with it and see what comes out...sure I'll have some more questions!!
Thanks Again!!
Tom
Does anyone know the specific technique to doing the reversible stretch?  I've been playing around with the mtf function in PixelMath with mixed results.  I'm using mtf(midPoint, $T) to stretch and mtf(1-midPoint, $T) where midPoint is set to the value obtained from an STF on the image (hit radioactive button, then wrench and take the middle value on the RGB/K line).  A test of stretch, unstretch, diff shows an almost exact match between the original image and the stretched/unstretched image.  Ha Before Stretch shows the pre-stretch setup while Ha After Unstretch shows the images after the unstretch and difference. 

Applying StarNet to the Ha images using the stretch and unstretch process noted above yields a pretty good result.  Doing the same thing with an Oiii image leaves stars and large white blocks in the image.  So clearly this stretch, starnet, unstretch seems to work well with high SNR images and not very well at all with lower SNR images (at least with this stretch / unstretch method).

Ha & Oiii Linear Starless & Oiii NonLinear Starless shows the original and starless images for Ha and Oiii using this stretch / unstretch approach.  It also has the results of doing a normal STF / HT permanent stretch on the Oiii followed by StarNet (which did a decent job of removing the stars).  So it would appear that the quality of the stretch is a key factor in the operation of StarNet.  Note both Ha and Oiii images had minimal processing - integration, DBE, and MLT (to remove the 1st wavelet layer).  Tests with earlier stages of the processing (no MLT and no DBE or MLT) showed the stars not being completely removed from the Ha image (this test using an STF/HT permanent stretch).

Any thoughts on alternative approaches to a reversible stretch that might produce better results with the lower SNR images?

Paul
 

Attachments

ac4lt

Member
Oct 8, 2018
5
0
I installed the 1.8.8-5 version of PixInsight yesterday and today I tried to add starnet++ back in but it's giving me grief trying to install (though I had no trouble last time).

I'm on macOS Catalina 10.15.3 and I'm putting the pixinsight files in the locations as specified in the readme. When I run install modules and click search a MacOS dialog pops up asking me if I want to move the dylib to the trash or cancel. I click cancel and then the button appears in system preferences to allow the library. I click allow and repeat the process.

This time I get a dialog asking me to cancel or open and I choose open. But then PI says that 0 modules were installed.

I must be doing something wrong but can't figure out what.

Anyone got any ideas?
 

pfile

PTeam Member
Nov 23, 2009
5,709
116
yes due to how -5 was built system integrity protection (SIP) is not allowing the module to load. it is unsigned because i don?t have an apple developer certificate and apple?s new security policy requires all dynamic libraries to be signed. juan says he will try signing the module and then i guess we?ll pass it to nikita to re-release on sourceforge.
 

eoverstreet

Member
Aug 4, 2015
16
1
I have the latest version of MAC OS, Mojave and cannot get Startnet+ to install either since the recent upgrade.  It installed with the prior version.  So I guess it's a MAC thing and not just Catalina