Starnet 2, is it actually not that good or am I doing something wrong? - Forums

Topic starter

Dec 26, 2023

So I finally bit the bullet an bought pixinsight and blur and noise xterminator (no budget for the star xterminator)
I went with the Starnet 2 alternative which seemed pretty good.

After using it on a few of my images I get really frustrating results with a lot of the star Halos still remain in the image.
When the image is streched, those halos really start to get in the way of a nice image.
I use the default settings for starnet and I'm really not happy with it. am I doing something wrong?

(I'm using it before I strech the image)

Before Starnet 2

After Starnet 2, you can see those dots all over the image, the images isn't streched far at all. An when the stars are bein put in, it really doesn't look clean at all.

Any recommendations or is it just the way it is?

Leonardo da Silva Garcia Leite

Dec 26, 2023 · Edited: Dec 26, 2023

That is the way it is. There is also starXterminator from Russell Croman, they have just minor performances differences.

I like to use it AFTER I have stretched the image (masked stretch usually). Otherwise, you will stretch the starless far beyond than the star_mask and it will make a poor merge when you combine them back. Even by re-stretching the star_mask will be difficult to match the intensities.

Helpful Insightful

andrea tasselli

Frequent contributor

Dec 26, 2023

It does a decent job I'd reckon. Are you screening the stars in and out? Also, I'd remove stars in the linear phase and then process them separately, if at all feasible.

Reg Pratt

Dec 26, 2023 · Edited: Dec 26, 2023

andrea tasselli:
It does a decent job I'd reckon. Are you screening the stars in and out? Also, I'd remove stars in the linear phase and then process them separately, if at all feasible.

This. Starnet2 has always worked great for me and there are no signs of its use when the stars are removed while the image is linear screened back on at the end. It's normal to see the black circles where the stars used to be in the starless image.

Removing the stars in linear form also has gives you control over how much you want the stars to be stretched. Basically a method of star reduction.

Brian Puhl

IOTD Staff

Dec 26, 2023

Looks about normal for Starnet2. StarX is far superior, but that's not to say it doesn't have it's quirks. StarX is more likely to eat filamentary structures in your data, so use it carefully. Starnet2 has always left behind a mottle in my work, and because of that, it's no longer in my processing workflow.

andrea tasselli

Frequent contributor

Dec 26, 2023

I just finished testing starnet++ 2.1.1 against SXT and starnet came out clearly on top on a linear image, with or without BXT applied beforehand. And thank goodness for that, too…

Wei-Hao Wang

Dec 27, 2023

As far as I know, StarNet++ is designed to work best on stretched images. I get very good results in the middle of the several stages of stretching.

Well Written

Jens

Topic starter

Dec 27, 2023

alright, so it sounds like I need to use it after streching. them image, that does make sense, since I don't see those artifacts when I don't strech it to far.
It's just a little frustrating when I captured enough data but I can't strech the image because of the artifacts Starnet 2 creates. I haven't had those issues with Starxterminator when I used the text version.

Die Launische Diva

Dec 27, 2023

I find StarNet 2 excellent, and in the rare case something is left behind, it can be managed with a retouching tool. Regarding your image, I believe the real problem is the lack of SNR for which StarNet 2 can't do anything about it, it just reveals the issue for you.

Happy Holidays and Clear Skies!

Well Written Insightful Respectful Concise Supportive

Kathy Walker

Dec 27, 2023

The original version of Starnet++ only worked on stretched images. The current version supposedly works on linear images, but I have never gotten good results that way. I only use it on stretched images. I get less ringing that way.

Well Written Concise

Nicola Beltraminelli

Dec 27, 2023

The result you have obtained looks good from my perspective. As mentioned by others, I usually do Starnet++ after image stretching.
My recommendation is always to verify the starless version versus the image prior Starnet++. As a matter of fact, it happens sometimes that thin filaments are deleted in the starless version.
Of note I saw a detailed comparison between Starnet and Starexterminator. In some cases Starnet is superior, in other cases it is the reverse. I use Starnet because is significantly faster.
CS,
Nicola

Helpful

Wei-Hao Wang

Dec 27, 2023

Kathy Walker:
The original version of Starnet++ only worked on stretched images. The current version supposedly works on linear images, but I have never gotten good results that way. I only use it on stretched images. I get less ringing that way.

I probably am still using the same 2.0 version. It does work well on linear images for me, but works excellently on somewhat stretched images. Do you mean the 2.1 version should work on linear images? That will be nice.

andrea tasselli

Frequent contributor

Dec 27, 2023

Wei-Hao Wang:
I probably am still using the same 2.0 version. It does work well on linear images for me, but works excellently on somewhat stretched images. Do you mean the 2.1 version should work on linear images? That will be nice.

Starnet++ 2.1.x works on linear images better than SXT in most circumstances.

LeePhillips

Dec 27, 2023

works perfect for me, I always stretch my images first in Maxim, then starnett them

christianhanke

Dec 27, 2023

A comment from my side. I´m using StarNet2 for a while. I´m always processing the background and the stars separately. The star picture is generated by substracting the starless picture from the original. By inspecting the areas between the stars the picture values are definitively zero. That means by generating the starless picture no artefacts are introduced. Only in the direct neighbourhood of the stars there are artefacts. By recombinig the two seperately processed pictures the result is in most cases satisfying.

Helpful

Jeff Kisslinger

Dec 28, 2023

Where are you guys downloading StarNet2? About a year ago it stopped updating in Pixinsight for me and when I did a search I read it didn’t work any longer. I just did a Google search and when I click on the StarNet website, it never opens. I liked having it because, as Nicola stated, some images work better with StarNet and others work better with SXT.

LeePhillips

Dec 28, 2023

Im not a Pix insight user, so Im using the M1 optimised version in terminal on my Mac, it's lightening fast. I altered the script so it produces the starless and the star mask .

Bogdan Borz

Dec 31, 2023

Hi @Jens,

Are you sure you're using the last Starnet 2? All this talk about linear does not make sense, it works perfectly well for linear images and it even has a tick box for selecting if the data is linear or not.

These spots are very unusual, they were visible for both SN2 or SXT in their first versions, now they work really well. I prefer SN2, you have interesting options with it (like 2x upsample). The problems are usually the very big, bright stars and their halos.

The problem I see with your image is that is very noisy. Sometimes this can influence how the AI is reconstructing (or making up) the blank space behind the stars. Clean data works much better since the background values are more uniform. Second, if this is an auto-stretch it will exaggerate those differences. I would check if it's the last Starnet version and try some light noise reduction before applying it.

All best,

Bogdan

Helpful