55 replies to this topic

[tenderfootnewbie]

I built an AI model for stretching astrophotography images!

 

You can try the model here: 

 

https://nebulamagic.com

 

Stretching images is one of the hardest things for people new to the hobby to learn.  While there are AI models for blur, noise, star, and gradient removal, this is the first AI model developed for stretching images.

 

Let me know what you think. It is free, and nothing needs to be installed. You can use it on your images.

 

Best regards,

-Doug

[Spaceman 56]

That is really cool Doug.

 

well done.  

 

Fitts files are big. can it handle them ok ?

[tenderfootnewbie]

That is really cool Doug.

 

well done.  

 

Fitts files are big. can it handle them ok ?

Awesome.  Good to hear!

 

Files up to 2G are fine.  I am working on compressing them on the fly to reduce the upload time and will release that when it is ready.

 

Best

-Doug

[rj144]

Thanks and nice work.

 

Just to let you know the download is a bit buggy.  And, it does a pretty good job, but the stretch is a bit on the conservative side for me and it doesn't do great for HDR type images.  Perhaps you could tweak that or have sliders to adjust that too.

[Spaceman 56]

Awesome.  Good to hear!

 

Files up to 2G are fine.  I am working on compressing them on the fly to reduce the upload time and will release that when it is ready.

 

Best

-Doug

thanks Doug.

 

to test your AI software system,  I am currently uploading a 300 Meg Fits file of NGC-6744

 

after selecting the file, and hitting upload, it  would be great to have some type of an indication box for how the upload is progressing. I am unsure if the upload is actually working right now, but I will give it time. 

 

I will also post the results if that is acceptable to you. 

Edited by Spaceman 56, 15 February 2025 - 03:34 PM.

[tenderfootnewbie]

thanks Doug.

 

to test your AI software system,  I am currently uploading a 300 Meg Fits file of NGC-6744

 

after selecting the file, and hitting upload, it  would be great to have some type of an indication box for how the upload is progressing. I am unsure if the upload is actually working right now, but I will give it time. 

 

I will also post the results if that is acceptable to you. 

There is a blue progress bar at the top of the page as the file uploads.  It is not very noticeable unless you know where to look.  300mb is fine. 

After the file upload is complete, it will stretch and display the image and you can compare it to the original.  There are some sliders below the image that allow tweaking of the results.  

Feel free to post the results!

 

Best

-Doug

[Lizardman]

Wow. Wonder though how long will it be before AI etc etc will completely suck the fun and challenge out of astrophotography? Ah , question for another day .

[tenderfootnewbie]

Thanks and nice work.

 

Just to let you know the download is a bit buggy.  And, it does a pretty good job, but the stretch is a bit on the conservative side for me and it doesn't do great for HDR type images.  Perhaps you could tweak that or have sliders to adjust that too.

Hmm.  Yeah, it looks like there is a delay between hitting "save" and the download start.  Thanks for finding it.  It should be a simple fix.

Also, thanks for the feedback on HDR images.  Sliders below the main image allow some limited tweaking of the model strength.  I have a new model in development with a larger training set, including narrowband images.  It should be a bit stronger.  

 

Best

-Doug

[RyanSem]

Wow. Wonder though how long will it be before AI etc etc will completely suck the fun and challenge out of astrophotography? Ah , question for another day .

Booo. This tool is hardly a problem. Are you going to complain about CMOS as well and claim we should still be shooting with film? Give me a break. 

 

Doug, this program is pretty neat. Did a nice job on my M81 image. I actually prefer the output to what GraXpert was giving me. 

[tenderfootnewbie]

Booo. This tool is hardly a problem. Are you going to complain about CMOS as well and claim we should still be shooting with film? Give me a break. 

 

Doug, this program is pretty neat. Did a nice job on my M81 image. I actually prefer the output to what GraXpert was giving me. 

Awesome!

 

Best

-Doug

[matt_astro_tx]

Neat!  Nice job Doug. I definitely think there's a space for this if you keep developing it into a standalone app.  Or perhaps one that can be integrated into programs the way starnet++ is.  I know I could use a hand with stretching from time to time!

 

This is great.

[w7ay]

I built an AI model for stretching astrophotography images!

 

Very nice, Doug.

 

I took an unstretched 230+ MB FITS image that has been gradient removed, NoiseXTerminated and BlurXTerminated in PixInsight, and sent it to your web app.  Without touching anything, I got this:

 

 

Everything popped up automagically, dust and all.  Perhaps just a tad over aggressive (the nebulosity lost some color).

 

I backed off the curve just slightly and boosted the saturation, also just slightly to recover some color -- spent less than half a minute (:-), and got this:

 

 

I really appreciate how it managed to bring out the dust with zero work on my part.  Very nice considering that you only think it is just "beta."

 

If it made is available as a native app on macos, or as a PixInsight plug-in, I can see myself using it as part of my workflow.

 

Chen

Edited by w7ay, 15 February 2025 - 06:00 PM.

[imtl]

One thing I don't understand and be glad to get an explanation.

Stretching is in a lot of sense an individual preference that leads the way. How does that sit with the fact that AI ''decides'' for me what is the best stretch?

[dcbrown73]

One thing I don't understand and be glad to get an explanation.

Stretching is in a lot of sense an individual preference that leads the way. How does that sit with the fact that AI ''decides'' for me what is the best stretch?

Processing images is pretty much to an individual preference.  I see a lot of oversaturated images.  They look pretty for sure, but do not come off as natural.  Some like that, some don't.  So, in the end, it's all preference.

 

That said, I think the idea here is to take the general consensus of what looks good and AI tries to manipulate the image to fit into a bell curve of that general consensus.

 

For instance, I have BlurXterminator and it has a default setting which is clear to me a lot of people use that default setting.  I find it a bit overly aggressive in my opinion and always dial mine back.

 

I haven't used tenderfootnewbie's new tool, but I'm excited about it.  I'm sure with time, it too may end up with more settable attributes to help conform to a specific users opinion or preferences, but you always have to start somewhere!

Edited by dcbrown73, 15 February 2025 - 06:19 PM.

[vidrazor]

They look pretty for sure, but do not come off as natural.

There isn't a single astrophotographic image, made by anyone, that is "natural".
 

[dcbrown73]

There isn't a single astrophotographic image, made by anyone, that is "natural".
 

Careful with blanket statements.  They generally cannot be proven beyond all doubt and are usually proven not true in the end.

 

I'm pretty sure most people know what I mean when I say, more natural looking.  We are on an Astronomy forum in the Astrophotography section.

[tenderfootnewbie]

One thing I don't understand and be glad to get an explanation.

Stretching is in a lot of sense an individual preference that leads the way. How does that sit with the fact that AI ''decides'' for me what is the best stretch?

dcbrown73 had a good answer.

That is why I made this model.  I am not very good at stretching.  It is fussy, time-consuming, and not very repeatable.  Every other part of the process is pretty mechanical.  

 

This model seeks to make a given image look like "what is published."  That is a better standard of what humans like than any math equation I could come up with.  

 

BTW, The system is designed so stretches are always "reversible," meaning there is no loss of information and nothing to prevent users from touching up the resulting images with other tools if desired.   

 

Best

-Doug

[tenderfootnewbie]

Very nice, Doug.

 

I took an unstretched 230+ MB FITS image that has been gradient removed, NoiseXTerminated and BlurXTerminated in PixInsight, and sent it to your web app.  Without touching anything, I got this:

 

Everything popped up automagically, dust and all.  Perhaps just a tad over aggressive (the nebulosity lost some color).

 

I backed off the curve just slightly and boosted the saturation, also just slightly to recover some color -- spent less than half a minute (:-), and got this:

 

I really appreciate how it managed to bring out the dust with zero work on my part.  Very nice considering that you only think it is just "beta."

 

If it made is available as a native app on macos, or as a PixInsight plug-in, I can see myself using it as part of my workflow.

 

Chen

Awesome! Great images, BTW, thanks for sharing them!

 

Best

-Doug

Edited by tenderfootnewbie, 15 February 2025 - 07:02 PM.

[vidrazor]

I'm pretty sure most people know what I mean when I say, more natural looking.  We are on an Astronomy forum in the Astrophotography section.

Well, now that you mention it, what do you mean? I certainly don't know.
 

[Exquisitus]

For fun I ran a master auto crop M42 through it.  Not bad results.  There were

no other processes, just the stacked file ran through your program. 

[dcbrown73]

Well, now that you mention it, what do you mean? I certainly don't know.
 

Yeah, you're just being argumentative.  That really doesn't provide any value to this thread.   

 

I will oblige you in an similarly overt way, but then I digress.

 

The left is a well know Astrophotography who does makes spectacular images, but in many cases.  The work is almost neon like and doesn't look real.

 

The right is my M31 from 2012.  It's a pure RGB image stretched with minor touch ups, but no overt changes.

 

[imtl]

dcbrown73 had a good answer.

That is why I made this model.  I am not very good at stretching.  It is fussy, time-consuming, and not very repeatable.  Every other part of the process is pretty mechanical.  

 

This model seeks to make a given image look like "what is published."  That is a better standard of what humans like than any math equation I could come up with.  

 

BTW, The system is designed so stretches are always "reversible," meaning there is no loss of information and nothing to prevent users from touching up the resulting images with other tools if desired.   

 

Best

-Doug

Hi Doug,

 

Between dcbrown73 and your replies, I fail to see a good answer in my view. There is no concensus on what looks good and there is no bell curve to fit to. If the goal is to get a none saturated image then I don't see a real need for AI here.

 

But in any case, I won't keep on poking at this in this thread. I feel my pondering will just cause people (not necessarily you) to respond in a defensive way and it will just ruin the topic. So, I'm out. 

 

Good luck with your script. I appreciate the work invested.

Edited by imtl, 15 February 2025 - 07:51 PM.

[imtl]

Yeah, you're just being argumentative.  That really doesn't provide any value to this thread.   

 

I will oblige you in an similarly overt way, but then I digress.

 

The left is a well know Astrophotography who does makes spectacular images, but in many cases.  The work is almost neon like and doesn't look real.

 

The right is my M31 from 2012.  It's a pure RGB image stretched with minor touch ups, but no overt changes.

 

You're kidding right? You just compared your image to one of the most iconic M31 Oiii arc images. There is no competition here. That thing actually partially did science and it is just spectacular and just as real as yours. The fact that your and my eyes cannot distinguish narrowband very well does not make anything here more real than another.

 

If anything, both are real and not real at the same time. No disrespect but if I need to choose an image like the M31 arc and your "real" image, I know what I'm choosing.

Edited by imtl, 15 February 2025 - 07:56 PM.

[dcbrown73]

lol.  It always works better when you follow the discussion before commenting.   I wasn't trying to compare my work to Bray's. (he is far superior to me in this hobby)  I was showing the difference between natural looking and clearly artistic creativity in an image.  (he asked!)

 

Anyhow, I'm out of here. I'm not looking to hijack tenderfootnewbie's thread.   This is about his fabulous work.  (sorry tenderfoot)

Edited by dcbrown73, 15 February 2025 - 08:04 PM.

[imtl]

lol.  It always works better when you follow the discussion before commenting.   I wasn't trying to compare my work to Bray's.  I was showing the difference between natural looking and clearly artistic creativity in an image.

 

Anyhow, I'm out of here. I'm not looking to hijack tenderfootnewbie's thread.   This is about his work.  (sorry tenderfoot)

I think I followed it very well. There is nothing more natural looking in your image than the other one. It's all in your mind. Enjoy your hobby. I don't see the need for all of this.

 

Back on topic.