FREE & Fully Automated Background Extraction - That does NOT use AI! Better than GXP?!
ฝัง
- เผยแพร่เมื่อ 6 ก.ค. 2024
- To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/cuivlazygeek . You’ll also get 20% off an annual premium subscription.
My Patreon: / cuivlazygeek
My Merch Store: cuiv.myspreadshop.com/
Seti Astro ADBE video: • AutoDBE: Throwing my h...
Seti Astro Scripts: www.setiastro.com/
GraXpert: www.graxpert.com/
ADBE Repo: raw.githubusercontent.com/set...
GXP Process Repo: pixinsight.deepskyforge.com/u...
Amazon affiliate: amzn.to/49XTx01
Agena affiliate: bit.ly/3Om0hNG
High Point Scientific affiliate: bit.ly/3lReu8R
First Light Optics affiliate: tinyurl.com/yxd2jkr2
All-Star Telescope affiliate: bit.ly/3SCgVbV
Astroshop.eu Affiliate: tinyurl.com/2vafkax8
-----------------------
My equipment
-----------------------
Dwarf 2 Smart Telescope: bit.ly/3SyChXua
Seestar S50: tinyurl.com/3n62hpzx
ZWO ASI2600MC Duo: bit.ly/3M8IkPS (Agena) or bit.ly/3Jti7KQ (HPS)
ZWO ASI585 MC Pro Camera: bit.ly/3Iwo6y8 (Agena) or bit.ly/4cp1Ofn (High Point Scientific)
TOUPTEK IMX571 APS-C Cooled camera: tinyurl.com/mr2yhvcw
TOUPTEK IMX585 Cooled camera: tinyurl.com/4y9dz774
OGMA versions (use code CUIV on checkout for a small discount!): getogma.com/
CarbonStar 6" Newtonian: bit.ly/3W7ZFiA
Mirror Mask for CarbonStar (designed by me :) ): www.thingiverse.com/thing:661...
SW Quattro 150P Budget Newt: bit.ly/3BABCxl
Backyard Universe upgrade kit: tinyurl.com/3juwm5sz
OCAL Collimator: s.click.aliexpress.com/e/_Dmu... or s.click.aliexpress.com/e/_Dc6...
Askar V Modular Telescope: bit.ly/3Imoq2M
or bit.ly/3GLuyR8 or tinyurl.com/2h758bvt
RedCat51: bit.ly/48hyuVx (Agena) or bit.ly/48pTWXW (HPS)
Cable Organizers/Focuser bracket: www.etsy.com/shop/SmartCompon...
WarpAstron WD-20 Mount: bit.ly/3wxNbGI
ZWO AM5 Mount: bit.ly/3Qb5xVf or bit.ly/3JeVUkn
ZWO EAF Focuser: bit.ly/45agP0Z (Agena) or bit.ly/4avCIcE (HPS)
Antlia ALP-T highspeed filter: bit.ly/42RBR2e
SVBony Filters: amzn.to/3yBuQZX
Mini Computer (newer version): amzn.to/49XTx01
Flocking paper/Felt paper: amzn.to/3zASgvu or amzn.to/3m94zwa
Follow me!
My Instagram: / cuivlazygeek
My Facebook page: / cuivlazygeek
This video was sponsored by Brilliant. - วิทยาศาสตร์และเทคโนโลยี
To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/cuivlazygeek . You’ll also get 20% off an annual premium subscription.
My Patreon: www.patreon.com/cuivlazygeek
My Merch Store: cuiv.myspreadshop.com/
Seti Astro ADBE video: th-cam.com/video/XqXfzr1ZLdk/w-d-xo.html
Seti Astro Scripts: www.setiastro.com/
GraXpert: www.graxpert.com/
ADBE Repo: raw.githubusercontent.com/setiastro/pixinsight-updates/main/
GXP Process Repo: pixinsight.deepskyforge.com/update/graxpert-process/ and
Amazon affiliate: amzn.to/49XTx01
Agena affiliate: bit.ly/3Om0hNG
High Point Scientific affiliate: bit.ly/3lReu8R
First Light Optics affiliate: tinyurl.com/yxd2jkr2
All-Star Telescope affiliate: bit.ly/3SCgVbV
Astroshop.eu Affiliate: tinyurl.com/2vafkax8
How is your comment 4 days old but your video some 20 minutes old
Seti Astro has a few amazing tools. Along with this there are 3 more I find indispensable - the statistical stretch (first pass stretching), rgb stars from narrowband data (no more r g b filters just for the stars) and star stretch (great tool for getting the star image ready to compose). All well worth a look
I need to check those RGB stars from NB data :) cheers!
Seti Astro has made a significant impact on the AP community. I've used his other scripts, and they work exceptionally well.
The best part? They're free!
Couldn't agree more!
Franks channel is my go to at the moment his repository of scripts are amazing. ADBE so much better thank GraXpert! Well explained Cuiv!
GXP does work very well in a pinch though, I think they're complementary tools :)
I was very pleased with the ADBE method! I always seemed to find a weird dark region around galaxies that were obviously not really there. Thanks Cuiv for your review. It’s really nice to see examples of different techniques and the results. Lots of TH-cam videos say what the result is but don’t offer anything to back it up. Clear skies!
I have struggled with gradients due to light pollution in my area, but these tools will help me greatly. I went back on some images that I never completed due to the severe gradients and they did a wonderful job.
Awesome video Cuiv! I think it is critical everyone checks the backgrounds they are extracting instead of going by faith on any of the tools they end up using! Now we all wait until MARS gets released and I can retire this script :)
The man, the legend! Hopefully MARS will work great, but I'm a bit worried because I simply haven't been able to make the Gradient Correction Tool work well for me... :/ fingers crossed!
@@CuivTheLazyGeek Gradient correction isn't really a great tool, but MARS will be perfect gradient removal. If you have tried Multiscale Gradient Removal, it will be like that but without the need to find reference data.
I found ADBE a few day before and was instantly very pleased. One of the advantages is not to load megabytes of models. I love this non-AI solution because of the elegance and really slim implementation. Congratulations to the Autor!
I'm using it for quite a while now (and his other tools) and I'm very pleased by the results.
Note that Gradient Descent is a common method used in AI/ML programming. The key insight here is that we can use Gradient Descent without the rest of what sometimes goes on with AI/ML programming, and this can give us a better overall result.
Another fine video with a great description!
Already added to my repositories. I plan to use both. Thanks for the links. Much appreciated.
I always try more than one background extraction when processing since they perform so different in my opinion on different targets and depending on the gradient in itself. That said this one looks to do a good job on all targets you tested it on!
That's the good way to get to it- compare and keep the best! :)
I've added most of Seti Astro's script to my workflow. Thanks for the video and side-by-side comparisons between Graxpert and automated BE. AI doesn't bother me since I know machine learning will improve with more and different samples added. However, the Automated BE does a fantastic job now, allowing me to stay lazy. Your videos are enjoyable to watch and I always learn something new.
He really has some great scripts!
Saw this on his channel, this was a great showcase
Both the standalone and Siril version of GraXpert actually have 4 different interpolation methods - RBF, Spline, Kriging and AI. I think the first three are mathematical while AI is obviously AI. In the first three you're able to create a sample grid or place and remove sample markers yourself so it is not completely automated like the AI method.
Thank god someone else noticed. There is nothing wrong with Graxpert and the AI algorithm is only one option.
Yep, I mention that around 2:24, albeit briefly
@@CuivTheLazyGeek ah, now that I rewatched it I understood what you meant, completely missed it the first time.
Honestly it's weird that they wouldn't include all the options in pixinsight version anyway but I guess it's important to clearly articulate that GraXpert is an amazing piece of software (and so is Siril) and people shouldn't think they have to spend extra money on Pixinsight and some extra paid plugins to start astrophotography. Realistically nowadays free software can get you 98% of the way there.
Cuiv thanks for the video. I've been using seti astro scripts for a while and they're great. Clear skies.
Good to hear! Clear skies!
Cool, thanks for pointing out this tool. It's always great to have many choices.
It really is!
Great video. Just a small comment. Gradient descent is the workhorse during neural network training. So people would label it as AI.
That's very true! but then it's just one of the methods used to train the model, so while it will determine the final weights of each layer/perceptron/whatever is being used, it's still just a mathematical thing that is absolutely not AI nor ML (otherwise I can say that a screwdriver I used to build my computer with is therefore a computer :p)
Yoooooo, keep up the great work!
Thanks! Will do!
This tool is brilliant! Better than Graxpert and anything else I have used so far. Thank you for the recommendation!
WOAH COOL!
As always, a Brilliant treasury of information for those of us limping along towards AP. I was unable to find a standalone version that I could open. I assume the non-pixinsight, stand-alone version is the last one on the SetiAstro list of scripts. I kept getting notification of a script error when trying to open the downloaded AutoDBE 'script'. Might be just as well to wait until I feel up to learning Pixinsight. cheers ... Gary
Purely coincidentally, I had installed the SETI add ons just before I found your latest vid Cuiv (at least this one of your tips isn't costing me money! 😂 ). Your suggestions about fine tuning are really useful, as I have been working on some weird artefacts on M16, though some of those may well be light leakage, and so far, my background extraction efforts have been less than stellar (pun intended).
I find it funny to say that it's not AI, it's purely mathematical :D
Anyway, you talked about the issue of local minima but there is also another thing : the minimum found could be signal, as in, dust cloud for instance. Such minimum should be considered the same as a local minimum I think. I don't know how the tool does thing so, maybe they take care of it. i wonder how it goes with tests though.
yeah gradient descent is a classic machine learning tool- the kind of machine learning that gets called "AI" by the media recently
I specifically included the User Defined Exclusion Area such that the user and exclude dark nebula (as I say in the instruction box) and show an example of that in my video as well. You are 100% correct it could think a dark nebula is background, and so does every other gradient correction tool. I wish they all had a means of excluding areas in the model!
Hahaha yes, ML is also purely mathematical, but more black box! And yes, you can exclude areas with the exclude function (I think I showed that in the video?)
I have see Franklin's teaser for the ADBE. I think the script is a good step forward. Nevertheless I am working into the Gradient Correction Tool. When the sliders are set correct the result is absolutely perfect. Sure you need two or three attempts for the best result but I think GCT is superior to Graxpert and also to DBE not matter is manual or automatic sample point setting.👍🏻
Using GCT in many cases you can even get away without any flat files. I have some good raw frames but no more Flats for them. GCT makes the data usable for further editing.
I was never able to get GCT to work for me - I watched the PI tutorials, read the manuals, etc. but nope very underwhelming results... Not sure what I'm doing wrong there
When Seti Astro released it I had a play and really like it
It's pretty good for sure!
I definitely think this could be a better tool than Graxpert. The torture test would be dark nebula like the Witch Head nebula, where the difference between subject and background is much smaller than with a galaxy or a huge blob of Ha.
Good point! Maybe I should have tested on some data from Telescope Live, since that nebula is hard to get from Tokyo!
Great comparison between the different tools. Seti Astro is on a roll at the moment, some really good tools that take the pain out of Pixinsight.
He sure is, amazing contributions!
When you did ABE function degree 2 on the gradient (at about 7 minutes), if the subtraction had been done right, what would you expect to see? A smooth image? All black, all white?
merci cuiv, je suis sa chaine et j'ai déjà installé ses tools, vraiment bien.
Excellents outils :)
The Multiscale All-Sky Reference Survey (MARS) is supposed to give us gradient removal based on actual sky photometry. For PixInsight users, will the need to any of these tools survive that release?
I hope not!
It will be great to see! But I haven't really managed to make Gradient Correction work for me for now...
Told ya 👍🏻
Hi Cuiv! I don't mind using AI tools but I feel that some of the training models are still in its infancy. Even though even Gradient correction is available in PixInsight, I still use ABE or DBE instead as I have a lot more control. I like this script that you have presented as it gives you a lazy way to use DBE 🙂 But you still have plenty of control. Thanks for the video!
Thanks Dave as always!
Hi Cuiv, thanks for this video. When you say that GraXpert doesn't give you much control about what it's doing, are you referring to the PixInsight plugin? Because the standalone version lets you edit the point grid, before extraction, by manually adding or removing points (just like Siril's "Background extraction" tool does)
I'm referring to the GraXpert AI tool only - I briefly mention the rest of the GraXpert methods as similar to DBE :)
I always found graxpert to be a bit aggressive but used it because it was much less effort than DBE, it seems that this script will be my preferred method from now on
Yep, exactly Ryan! Usually with a smoothing of 1 GraXpert is OK, but sometimes even then too aggressive
I do hope Siril keeps up with pixinsight for the poor!
Very good indeed, my only problem is Siril do not respect pixels outside of actual original frame in registered frames, when staking, it weight same (must be ignored if outside of original frame) and final image looks like sht especially if have big field rotation. I have no tracker, it is very annoying.
You can change the stacking to be maximum though?
I hope so too!
I forgot to ask: Have you already compared the new GradientCorrection vs. ADBE?
Great video Cuiv! I'll try it just as soon as I can get the repo to install (get the error: Unknown code signing identity 'FranklinMarek'). Probably just me!
You'll need to upgrade to the latest version of PixInsight, or the latest subversion (the 4 digits at the end of the version number) and it will work!
@@CuivTheLazyGeek Yup, been putting this (upgrade) off... now I have no excuse!
Cuiv! This is a fantastic tutorial & evaluation of the plug-in! As a retired teacher, I greatly appreciate & admire your analogies (.e..g finding the base of the mountain, etc.) and the examples you use to establish & reinforce understanding & applications! As always, I will 'bookmark' this one. Kudos! 👍👍
FWIW, I do have a few questions (apologies ahead of time if the answers should be common knowledge...I am a relatively inexperienced 'lazy geek'!)
1. Did you apply darks/bias/flats/standardized stretching to all three of your images before applying the various methods of background extraction?
2. Is it generally better to do background extractions on LRGBHaOIIISII channels individually, then combine the results, or after the final colour/false-colour image has been assembled?
3. Is there a quantitative rule of thumb to identify situations when any attempt at any level of background extraction will do more harm than good to the image (e.g. imaging with a well-baffled OTA, from a truly dark site, without effects of the moon, zodiacal light, etc....and well-executed flats...or will background extraction still be helpful?)
Thanks! 👍
Thanks so much Derek! To answer your questions:
1- bias/flats applied (no darks, I explain why in another video), no stretching (just display only auto stretch)
2- Yes if you have separate channels at first, I think that makes sense. It may even be better to split HaOIII and SII OIII OSC channels and do the BE there, but I'm usually too lazy!
3- I don't think there is a situation when ANY BE will do more harm than good - you can always judiciously choose a BE method that will improve the image (if only ABE of degree one or zero)
Cuiv, did you crop the images prior running Graxpert? Just asking because it doesnt work properly without cropping
Yes, quite a generous crop at that, but maybe I should have been even more generous with the crop?
@@CuivTheLazyGeek No, I think as long as you crop a few pixels it should be okay. I just thought you may have forgotten that. Usually Graxpert should get rid of the gradients easily, especially those on M31 confused me somewhat.
Did the Grx photo get cropped BEFORE processing? Please and thanks
Yes quite a generous crop was applied first
👍👍👍👍👍👍👍
Would it be cheating too much to blend the best of the nebula/galaxy from the original images with the fixed background images in photoshop or something? Perhaps not with the cone nebula, but with the others. This might be an unpopular opinion (or it might not), but I couldn't help but think the original images were best in basically all three scenarios. Yes, there was obviously noticeable gradients, but the nebula/galaxy always seemed to look its most beautiful and smooth and nice in those original images. And I'm not sure losing that is worth fixing the background.
Will this work with Affinity Photo?
I find Affinity Photos own background extraction tool works pretty well anyway. I bought Gradient Exterminator and to be fair I didn't find it that much better, in fact I find for some images the Affinity tool is actually better as you have more control.
Hi Cuiv. Dosen't seem to like the latest version of PixInsight (v. 1.8.9_3) Do you know of any issue?
Can ADBE be used standalone without the need for PixInsight?
no as it is a script that incorporates ABE and DBE within pixinsight
@@nikaxstrophotography Thanks, so it is not better than Graxpert then, plus I noticed that Cuiv seemed only to focus on Graxperts AI function with extreme high or low settings.
Didn't seem like an objective test to me.
Graxpert is free, ADBE costs €300 since PI is required.
Unfortunately not - re the test, not sure how to make it more objective, would need to test each gradation of the slider?
@@CuivTheLazyGeek Thanks for responding . I think the problem is that everyone is jumping on the "AI" bandwagon and assuming it is the best. I get sick of hearing about it in every walk of life.
I use Graxpert as a standalone tool in my workflow with Siril. Yes, I sometimes use the AI algorithm, usually at a high setting, but I will also often use one of the three manual algorithms within Graxpert - whichever to achieve the best result for my specific image.
In your review, you focused entirely on what Graxpert's AI algorithm does, but that is only one of several options that were completely ignored in this video.
In addition, you didn't do any cropping to remove stacking artifacts before processing - everyone I have ever heard from always recommends this as one of the most basic steps to be done before any other calibration, hence the reason why it is the first step in the Graxpert (standalone) process.
I would have preferred to see "Pixinsight Required" in the video title, in which case I would not have watched it.
My PixInsight won’t allow the download because of invalid signature from seti astro.
Make sure to have the latest version (including 4 digits subversion) of Pix insight!
@@CuivTheLazyGeek that worked. I forgot I didn’t download the last version because it kept getting hung up on the ref doc. I tried it again today and it downloaded. Now I’ve got the seti astro scripts. Thank you so much!
Note sure how anything that requires a very expensive piece of software can be described as "free"
The plugin created by SetiAstro is free. GraXpert is not free, whether you’re using it with PixInsight or Siril.
btw i send you message on messenger about Polish harmonic mount. I have hope you make video about it. Just like its creator, whom I informed about it after the fact
I saw! I can't buy and review every mount I see though :( it does look very interesting of course :)
@@CuivTheLazyGeek I know, I'm sure you have many suggestions. But I hope you will find the time and funds to test it someday. I recommended it selflessly and without any expectations, out of respect for your work
Okay.
Nice video but you start off as if talking to someone new to astrophotography then continue with a plugin for a expensive piece of software. If you removed all the plugins for Pixinsight I wonder if it would still seem worth the money?
The plugins make using PI easier and more comprehensive, but I guess that's a natural progression with software. Without the plugins I still think that PI is worth the cost, although I recognise that there are other alternatives (free or otherwise).
AI is only as good as the training data. It will improve, but if there is a good purely mathematical way to get there, I will probably always opt for that. I use lots of AI tools, so I am not at all anti-AI. I thinks it's got great potential, but math is math and not an educated guess based on a statistical analysis of many thousands of processed images. Yes, a lot of great math goes into creating the AI model, but the outcome is still an educated guess based on training data.
I like being able to know exactly what the tool is doing - a big theme of Stanley Kubrick's 2001 is Man losing control of its tools (after gaining it in the movie intro) and the pure ML stuff is exactly that - you never know HOW the ML process arrived where it did :)
Interesting, but I'm not buying Pixinsight any time soon.
"AI" vs "machine learning" vs "purely mathematics" is a hazy vague distinction.