StarNet ++ and Windows 10

Carl O Beirnes

Carl O Beirnes

New member
Hi all,

Forgive me is this was already posted somewhere. I'm just wondering could anyone could help me on how to install StarNet++ onto PI. I'm using a windows 10 PC i'm very new to PI and haven't a clue where to start. Thanks for any advice and help.

Clear Skies,

Carl.
 
Download the relevant installation files from here, which includes installation instructions:

Best Open Source Desktop Environment Software 2024

Compare the best free open source Desktop Environment Software at SourceForge. Free, secure and fast Desktop Environment Software downloads from the largest Open Source applications and software directory
sourceforge.net sourceforge.net
 
Starnet DOES NOT remove stars, it removes bright pixels from the image without any care
are these small galaxies, asteroids etc. Starnet does not access the Gaia catalogue to distinguish
between stars and other objects. The only tool to create a "real starmask" out of the GAIA catalogue is
the script "MaskGen" in PixInsight. Everything else (including PI Starmask tool) is simply destroying every small bright object.
So be careful when using the so called Starnet, especially in areas where hundreds of small PGC Galaxies are in,
they will be erased. So Starnet should be better called
"bright pixel removal tool".

Gerald
 
the neural net was trained on starless images and their corresponding star-ful images. while of course this can result in false positives depending on how high-quality the starless images were, i think to characterize it as a "bright pixel removal tool" is completely unfair.

it will certainly mistake some small, faint galaxies for stars, but there's nothing in the code per-se that looks for bright pixels. it's trying to recognize star-like objects and inpaint them.
 
pfile said:
the neural net was trained on starless images and their corresponding star-ful images. while of course this can result in false positives depending on how high-quality the starless images were, i think to characterize it as a "bright pixel removal tool" is completely unfair.

it will certainly mistake some small, faint galaxies for stars, but there's nothing in the code per-se that looks for bright pixels. it's trying to recognize star-like objects and inpaint them.
Click to expand...
But...and I see his point, if the galaxies look like stars, they will be removed as well. However, my understanding is that after one removes "bright pixels" and processes the image, the "stars" are put back. So the galaxies would be put back too.

By the way--what is teh best way to make a perfectly fitting starmask in PUI that covers all stars from tiny ones to large ones?
 
rdryfoos said:
But...and I see his point, if the galaxies look like stars, they will be removed as well. However, my understanding is that after one removes "bright pixels" and processes the image, the "stars" are put back. So the galaxies would be put back too.

By the way--what is teh best way to make a perfectly fitting starmask in PUI that covers all stars from tiny ones to large ones?
Click to expand...
This is the way I use StarNet as well, and generally to get a good star mask subtract the starless image from the original image
 
The best way to make a perfec fitting starmask in PI is the PI script "MaskGen" because it accesses the GAIA catalogue
for creating the starmask. And it creates all stars from tiny to large ones. PI StarMask process and Starnet cannot do this! As long the so called "Starnet" does not access the GAIA or similar catalogue its fair to call it (together with PI Starmask process) a "bright pixel removal tool". Especially in areas where many many tiny galaxies are located the removal of them is contraproductive. And there are many areas with galaxy clusters with hundreds of it on one image!
So, i wanted to create awareness that the name Starnet is misleading and the use could destroy the presentation of galaxy clusters and does not improve them. To combine it with MaskGen gets you on the more real and not "PS" side...
Gerald
 
for everyday use i have not found a better tool for star removal than StarNet++. by everyday use, i'm talking about images taken from average to poor seeing with small instruments. i don't think any of my images run the risk of resolving tiny galaxies.

further, for the most part i am using this with images of emission nebula and there's no background galaxies to be found there, at least with the targets i am shooting with my short FL refractor.

if you are working with data from large scopes and dark skies then the "galaxy eater" problem might be a real issue. but i have certainly seen StarNet ignore things that are small yet don't have a stellar profile. honestly i have never tried to use it seriously on non-emission nebula images because it never made that much sense to me. and like rodd and dave said if you put the stars back you'll also put any deleted galaxies back as well.

anyway if you don't like what StarNet does there is nothing stopping you from retraining the network with your own images.
 
There are several way to use Starnet++
You can use it to just make a nice image without stars (with plus or minus good results) : ok why not
I use it to combine for example Ha signal (starless) inside RGB image with stars. Or to boost SHO image with 1 or 2 narrowband signal (starless too). No need a highest quality for this.


Anyway : some solution for PI OSX Starnet integration ? I saw some discussion about signing libraries.
 
As a PI newbie, although I am beginning to get it, I find all this information extremently useful. It is helpful to know in advance that this tool might remove your tiny galaxies and also good to know that it can be used very effectively to remove stars in the application of LFHH and to make an effective star mask. Always good to be in the know before wasting time on the wrong path.
 
Yes, amacfarlane!
One of the informations i wanted to spread is also that the PI Script
"MaskGen" allows to create real starmasks based on the GAIA catalogue database.
I personally also use Starnet sometimes, but only in conjunction with MaskGen to be on the safe side.
Gerald
 
oldwexi said:
Yes, amacfarlane!
One of the informations i wanted to spread is also that the PI Script
"MaskGen" allows to create real starmasks based on the GAIA catalogue database.
I personally also use Starnet sometimes, but only in conjunction with MaskGen to be on the safe side.
Gerald
Click to expand...
Thanks for pointing out MaskGen....sounds like the better script!
 
MaskGen uses catalog data to create a starmask, not the information from your picture.
So it needs to be configured to fit your picture, has limited magnitude,
and sometimes its hard to finetune the resulting mask to your existing starforms.
But once you get a little used to it, it is very precise and convenient way to get a really proper starmask.

Starnet is good in erasing stars in NB pics especially, and you wont use NB for imaging galaxy clusters.
It is also good in filling in the erased stars with context from your picture, something a star mask will not do by itself.
So take the advantage of all possibilities, and try to avoid the problems. :)

When downloading Starnet, there is a comprehensive readme on how to install.
If there is a problem doing that procedure, its best report also on that problem.
That helps focussing on that particular issue, instead of giving general hints.

Tommy
 
One thing that may have gone unnoticed is that if you make a starless image (starless) and subtract it from the original image to get the stars (stars) and next alter the hue, tone, brightness of starless, using restarred = starless' + stars does not get the correct star colors back in the restarred image. There are complicated ways to come close to doing that, but the only thing I can figure is to substitute the stars from the original image into the processed starless image (using the image stars as the transfer mask). Does anyone have a better method?
 
You must log in or register to reply here.
Back
Top