Revisiting Topaz Denoise for Astrophotography! vs GraXpert and NoiseXterminator!
ฝัง
- เผยแพร่เมื่อ 3 พ.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/
Topaz Photo AI: www.topazlabs.com/topaz-photo-ai
Graxpert: www.graxpert.com/
NoiseXterminator: www.rc-astro.com/software/nxt/
Amazon affiliate: amzn.to/49XTx01
Agena affiliate: bit.ly/3Om0hNG
High Point Scientific affiliate: bit.ly/3lReu8R
All-Star Telescope affiliate: bit.ly/3SCgVbV
Astroshop.eu Affiliate: tinyurl.com/2vafkax8
Dwarf 2 Smart Telescope: bit.ly/3SyChXua
Seestar S50: tinyurl.com/3n62hpzx
-----------------------
My equipment
-----------------------
ZWO Seestar S50: tinyurl.com/3n62hpzx
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/
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
ZWO AM5 Mount: bit.ly/3Qb5xVf or bit.ly/3JeVUkn
iOptron CEM70: bit.ly/431Ii2U (successor of my mount)
MiniGuideScope: bit.ly/3MyUJhn
Antlia ALP-T highspeed filter: bit.ly/42RBR2e
Celestron C6: bit.ly/3ocePVs
Hyperstar C6: starizona.com/products/hypers...
Mini Computer (newer version): amzn.to/49XTx01
AstroPC Pro: astro-gadget.net/gadgets/astr...
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/
Topaz Photo AI: www.topazlabs.com/topaz-photo-ai
Graxpert: www.graxpert.com/
NoiseXterminator: www.rc-astro.com/software/nxt/
Amazon affiliate: amzn.to/49XTx01
Agena affiliate: bit.ly/3Om0hNG
High Point Scientific affiliate: bit.ly/3lReu8R
All-Star Telescope affiliate: bit.ly/3SCgVbV
Astroshop.eu Affiliate: tinyurl.com/2vafkax8
Dwarf 2 Smart Telescope: bit.ly/3SyChXua
Seestar S50: tinyurl.com/3n62hpzx
Hi Cuiv, you might want to check out the new GraXpert Denoising AI model v2.0.0 which we released just a few hours ago. It has to be used with the new GraXpert version v3.0.2. The new AI model performs better on dark nebulae, produces less artifacts and solves some issues with denoising in brighter objects. As you pointed out, GraXpert is best used in linear state directly after background extraction :) In the linear state, the information GraXpert uses to denoise the image is much less compressed to small pixel value ranges, therefore GraXpert can produce better and sharper results in linear state
I love your Graxpert software!! Thank you for making it available to everyone for free. It has catapulted my images to a new level.
That's awesome!! Thanks, will do!
Maybe you should mention that Topaz is most likely not trained on Astro pics and might from time to time produce artifacts which don’t belong there. GraXpert and NoiseXterminator are trained with Astro only, which should deliver a better fitting Model not introducing non-existing details, which might be helpful in cases Topaz is designed for - denoising „real world“ earthly pics like wildlife & landscape etc. to let them look more realistic. GraXpert and maybe also noiseXterminator got a more scientific approach concerning details. Big difference.
Good point, and absolutely right! I actually had a whole filmed section on that (and BXT logic) but it made the video much heavier and I cut it out...
Release it !
Yes and no, I am using topaz for few years had all Denoise, gigapixel, sharpen ai, photo Ai and my first love is bird photography and from experience in my humble opinion topaz is great but it's over sharpening is not only happening to astro photos even using my R5, R6 and r7 shooting birds in the face at short distance 20feet and get ultra good feather details in raw photos, default ( ai autopilot) always butchers the photos inventing details making them look like plastic mold 😂 don't get me wrong is fantastic software but autopilot is way to pushy on sharpening inventing new details.
I believe it's done across all genres of photography.
Still amazing tools like magic just as all things today too much stareoids 😂😂
Informative and interesting. I enjoy the learning process in this hobby. Thank you very much!
My pleasure!
I just found your channel a couple days ago when I was looking into ZWO cameras on the used market. I am not yet doing astro photography but your videos demystify the subject and it's fascinating. I think your enthusiasm is fantastic. Your presentation is great!
Graxpert tip for OSC fully processed image. Open the fully processed image in Graxpert then apply a 10% stretch which will take it to an almost Hubble or SHO palette. Then save the stretched and processed image. You will then need to open it in Siril and save it as a tiff file. The image goes from a reddish or HHO to a colorful masterpiece. I stumbled on this when trying to remove a persistent gradient on a fully processed image. If you don’t open it in Siril and just try to save it in Graxpert it will be overblown and unusable but when opened in Siril it will look the same as it did in Graxpert. I have used this method on several of my mages and it looks amazing.
Wait whaaaat?? Can you share the exact steps? This is for HOO OSC images?
Is the below correct?
1 Process and stretch image as normal HOO, end up with very red image
2 open the image in GraXpert, stretch 10%, this magically makes it into SHO palette
3 Save image as tif and open in Siril
4 Save as tif from Siril again
Is that correct??
Yes, that’s what I stumbled upon while trying to remove a stubborn gradient in a fully processed image. 1. Load image in Graxpert
2. Click on 10% stretch at the bottom 3. Save stretched and processed. 4. Open the saved image in Siril. 5. Save in Siril as 16 bit tiff
If you don’t open it in Siril then save it in Siril it will be over baked and will be unusable.
I just tried it again on one of my processed images and it looks great. Check out my Astrobin images of the Lagoon Nebula. /user/CosmicCoywboy
I also used this on my images of the PacMan, Rosette, Fish Head, Eagle and Tadpole posted on astrobin. The images were all red before the Graxpert stretch.
I just found that if you save it in Graxpert as a tiff you don’t need to do the Siril steps. I was saving it as a32 bit fit before. So now it is super simple.
That’s 16 bit tiff. A little color enhancement will he needed but that’s easy once this is done. I don’t understand how or why this works but it does at least on my nebula photos.
Training is what defines how well an AI algo does on a dataset and I feel Russ did a great job with training his AI models. This also means that all AI tools will have some nuance when it comes to performance. This was a great example between some common tools and give us better insight into the cons and pros. Thanks for doing this! Clear skies!
My pleasure Dave thanks as always for your support!
Great comparative analysis.
I have just finished experimenting with the new Graxpert and let me say, it is awesome. I reprocessed my image of the Lagoon Nebula by stacking and star extracting in Siril then I stretched the image in Graxpert. Yes, you heard me right! I used the 20% stretch then background extraction and denoise and then unlinked the channels before saving. Recomposited stars in Siril then a little brush up in Affinity Photo. It took less than 10 minutes, I am pumped! Graxpert did a fantastic job and it retained all the fine detail in the image. I had no idea it would turn out so good.
Another note in Topaz denoise/ new photo AI you have to adjust your sliders they are there for a reason not just click and done approach.
Love these comparison videos! Tho one tip I'd like to offer is to upscale your videos to 2160p or at least 1440p before uploading (even if you record in 1080p). That's because 2160p and 1440p allow for much higher bitrates and therefore will preserve WAY more detail than 1080p making the comparisons clearer.
Good idea! I'll try that next time!
Of all of the comparisons of the new denoising tools I’ve seen, this is by far the best and most thorough. Thank you for this gift! You definitely just saved me $200. 😂
Glad it was helpful and saved you money!
Excellent, thank you. I use NXT and Topaz, glad I have the “old” version now… which one I use just depends on the image, most get a light treatment of NXT with very low settings to start with. Then more later in the process if needed. When I use Topaz it’s usually as a touch up to a final image.
At least with the old Topaz you can indeed control how much detail it adds (or doesn't!)!
I have been using Topaz Denoise since it came out and I agree with your assessment that it improved over time. Still using it and like that it is integrated with Photoshop. Also tried out GraXpert, which seems to work well. Don't have Noise Xterminator,
Thanks for the updates...always anticipated!
Yep, I'm happy the latest version is so good, unhappy the new Photo AI is so aggressive :)
Personally I find StarTool's Denoise (along with its other available tools) to be much more deft in removing noise without adversely affecting stars or object structure.
I just tested on an old m51 image took with a C8 and ASI183MCpro 5h worth from Bortle 8-9, I dont use noiseXterminator, but always used multiple sections of Topaz Denoise AI, I really got a good result by starting with GraXpert denoise at 75% and moving on to Topaz Denoise AI. I even added a fair bit of sharpnes at the end as I am not as purist as you are, but I have overcooked sharpneing... the denoise sharpening actually feels very natural and not overcooked, even tho I know it is really inventing details and not deconvoluting... mixing both is really powerfull one thing that I hate about Topaz denoise AI is how the BG becomes like lots of blobs, and starting with GraXpert totally made it go away. You should test it later on. I will definitely use this as standard workflow now.
by the way I used the 2.0.0 denoise algorithmon v3.0.2
I wrote in the previous video that I like Topaz better, but I use the old version, besides, after his work I still work with masks in Photoshop, precisely in order to minimize some of the artifacts of his work, in my opinion there are no worse, better noise editors , it’s just that each specific image obtained on each specific matrix requires individual processing technology and software.
You read my mind! I already have / use Topaz DeNoise and find it works fairly well - I was curious how the two compared.
Glad this is helpful!
I purchased Topaz labs at the behest of another you tuber and found it severely wanting on astrophotos. Fool me once shame on you(not you of course)…. I do have the latest version of denoising and have not purchased the new photoAI. I do like noisexterminator and have also used Graxpert where I haven’t found it entirely useful yet. But all of these will continue to evolve over time. My expectation is that NoiseXterminator will continue to be good for Astro.
Yeah it has its limits for sure... Agree on NXT!
@CuivTheLazyGeek, I have a quick question for you. I've had a number of refractors and I'm more and more convinced that I want to try a newtonian telescope next. Would you still recommend your Skywatcher Quattro with all the upgrades and pain to get it running properly, or would you recommend something else of similar size for a little more money up front? Thanks!
Right now I'd recommend the Apertura CarbonStar - it's basically a Quattro 150P but with none of the headaches (review coming next week)
@@CuivTheLazyGeek Awesome, thanks for the lightning fast reply :)
Tiff seems to be a prefer format, can these tool use raw?
I'm wondering... since software can easily determine much finer grey scale levels than the human eye can, is TPAI actually Inventing structure that isn't there, OR is its perimeter to separate out light vs. dark set at a level that's showing us very fine structure that we weren't previously aware of?
I Guess you'd have to run it a few times at the same settings to tell if the AI was being creative by altering the structures differently during each run, OR if its perimeters are more exacting and pulling out more details. If its the latter, it just boils down to aesthetics, preferring one over the other. Thanks!
To unveil existing structures that are embedded in the image, applying a mathematical deconvolution is what works - that's what BlurXterminator does. With TPAI you can compare with images from the Hubble and see that the details that TPAI added simply aren't on the Hubble pic
@@CuivTheLazyGeek Thanks for the info! I've been thinking (We're in trouble now! 😄 ) so...
I DL'd PIA15416 (M31 @ 9400x7000px) from images.nasa.gov and approximated your zoom, and most of the 'cloudiness' in your image where your seeing structures is stripped away. So of course there are no structures to see.
As I understand it a "mathematical deconvolution" is still image processing using computers which makes decisions using a man made algorithm(?) based upon humans choices (which may or may not be AI related), and in NASA's case also using decisions and tweaks by the imaging people before its released to the public. IOW both computer and human interpretation in the publicly released image.
Unless you have the original unaltered NASA photos to compare to, it seems to me we still really don't know of those wispy structures are actually there or not.
Then there's Apophenia, which is basically faulty pattern recognition inherent to all humans, to consider. Exaggerating slight differences that may or may not be there.
OR maybe my 66 year old eyes and brain aren't what they used to be! 😎
Hello, compare it with Darktable Astrodeoise and in GIMP with the G'MIC-QT filters... you can also use masks wonderfully.
Ooh another tool I didn't know about!
Thanks.
Cuiv, did you try moving the slider all the way to 100% in Topaz Photo AI? Does this effectively turn off all sharpening to the image just as was possible with Topaz Denoise?
Yes! It turns off all the sharpening - but in effect the noise reduction also is almost fully turned off...
I use Topaz Denoise with Starnet gui and GIMP. I don't mind a bit of detail adding but it is perfectly scalable (without the stars), wish they would do a special astro version. Tried Graxpert but the saved images are vastly different to those on screen. The built in GIMP tools are very useful like high pass for sharpening nebulous edges and screen mode on a stacked layer to easily create a stars layer. I wish someone would cook up an astro tool set for it as well, just to make it a little easier. Particularly for high dynamic range images, like the Orion nebula. With others you can throw in 3 or so image stacks of different exposure and they will blend to preserve detail, GIMP can do it, but not easily.
Apparently if you open the saved images with Siril and then save again, they can open fine in Gimp!
I have all those tools. RC Astros tools are best overall. I do use Topaz"s AI after I create a JPEG out of a TIFF to upscale an image without noise reduction or sharpening enabled. I take the result into PS and use a highpass filter to do additional sharpening.
That sounds like a very good workflow!
NoiseXTerminator is my go to I think it does the best job in denoising at the same time preserving details.
Yep, I agree! It truly is a great middle ground
Which version of GraXpert did you use 3.0.0, 3.0.1, or 3.0.2? Nice comparison.
Not sure right now
Will stay with GraXpert Denoise- Love it and it´s for free :)
Free is good :)
Good morning,
Why aren't we using star databases to "clean up" our photos?
I mean, we take a picture or many preferably of the sky. We stack them.
What do we see? Stars hopefully, but most of the bright pixels are noise, faulty pixels, atmospheric effects, dust, hou name it.
So why not just use the star databases to identify which bright pixels are actually stars and simply subtract everything else?
Been recommending topaz for a long time the best denoise for astro by a good margin especially for color noise. It’s so much more adjustable and effective.
I'm very wary of it - Topaz Denoise, now discontinued, seems to be fine with the right settings, but Photo AI really insists on creating details that don't exist (and sometimes not the same details on the same image)
@@CuivTheLazyGeek I don’t use photo AI I only use the denoise tool and it doesn’t creat any artifacts or introduce any additional data. I don’t use the sharpening just the denoise portion and it performs really well. I simply won’t upgrade and will just continue to use the Denoise tool on its own. I have topaz sharpening as well and it’s pretty good again with the right settings you have to use the ones that give you a nice soft subtle effect and then it’s very useful. Sometimes it’s more about knowing how to use the tool itself and what settings work best for different types of images.
Based on what you’re saying glad I never upgraded to the full photo AI suite and just stick with the tools I needed.
Have you tried Astrosharp ?
Hi, I’ve been using Topaz AI for a while but was a little worried about the ‘added details’ you mention. Gave Graxpert 3.0 a go and it seemed ok but it intermittently gives me ‘not responding’. Sometimes I get it to work, sometimes not. Unfortunately I’d say mostly not working. Topaz works every time. Think I’ll be sticking to Tppaz unless someone has a solution to my problems with Graxpert.
I hope you can get your GraXpert issues fixed .
Awesome..input reality and output artificial reality. It's all the magic you have left, Cuiv. Here's some plastic flowers for you 🎉
Thanks, I love plastic flowers, they're long lasting too!
@@CuivTheLazyGeek awesome..I'll let you go back to pretending it's honest to push software that takes astrophotos and alters them to SHOW WHATS NOT THERE..not like there is a difference in what is real or not to you anymore..as long as you get paid 💐
I'd say that deepsnr is currently the best denoiser, even if its quite picky with the data its fed
What is Deep SNR ?
@@tomhoskins4913 denoising ai made by the same guy as starnet, currently only a module in pixinsight but should soon be integrated into siril as well
Need to have a look, first I've heard of it!
@@CuivTheLazyGeek I suggest lookin up "DeepSNR tutorial" here on youtube and watch the video by zegery, helps a lot when first introduced to it especially since it's picky.
My experience agrees with yours. The Topaz stuff leans heavily into inventive AI. It was trained on terrestrial images as well, so you can get things like feathers on galaxies at extreme settings of their old Sharpen AI. When you run at lower settings it is still inventive, but folks often don't notice it because it's at a much smaller level. It's a bit like regulating the acceptable level of rat poo in your breakfast cereal. If they *did* train on astro, it would still be inventive, but you wouldn't notice it as much. I don't recommend Topaz for astro at all. And knowing how it works, I also don't recommend it for other images where you care about any level of "truth". It even made me *highly* skeptical of BlurX when it first came out because I had such a bad experience with Topaz (and things like AI phone photo sharpening that turned a nighttime photo of a parking lot into a lawn).
If folks *do* use tools like Topaz AI, they should disclose its use in their image info.
Yep, I agree here - I'm usually able to spot Topaz-processed images fairly quickly... BXT has a much stricter approach to things..
It's really hard to use these handy tools without an imaging telescope like the CARBON STAR 150 to get raw data from!
Well, to be honest the M31 was taken with Redcat+Askar V, and the Heart was taken by the quattro 150P :) but the CarbonStar is good :)
@@CuivTheLazyGeek I know... I'm 100% just giving you a hard time for the CARBON STAR 150 review!
If it isn’t broken, don’t fix it. Both GraXpert Denoise and NoiseXterminator work well AND are integrated into Pixinsight. Unless you want to leave Pixinsight and edit in Photoshop, I see no rationale for Topaz denoise. On the other hand, if you are wedded to Photoshop (or for non-Astro photos), Topaz Denoise is a good option.
Agreed!
In my experience Topaz Denoise creates nebulosity artifacts and also destroys some nebulosity too. I will stick to the existing astro tools.
Yep, I fully agree
It makes me chuckle when you comment on a Program adding information that you think shouldn’t be there then go ahead and process the Image to a degree that satisfies just you . Most of the Images I see online look like candidates for Paint-by-numbers . LOL ! As long as you’re not being paid by NASA for an image then all it needs to do is satisfy yourself with the final image .
Cuiv, I posted a quick video on youtube titled Graxpert 10% stretch trick. Please forgive the awkwardness of the video. I am not a youtuber and don’t want to be.
I saw! I'm going to test it out and might feature that on the channel if you don't mind!
I don't mind at all! Anything to help my fellow Astrophotographers! Love your channel! You are the reason I started doing this last year! Keep up the great work!
Good point…AI can hallucinate…
There is not anything close to NxT. Best purchase ever to make my photos better.
The RC Astro tools are incredible
i have to disagree here. It doesn't look, to me, like its creating detail out of thin air. It appears to me that its enhancing the blurred image and bring out detail that IS there but slightly hidden
No, I can get it to create different details with different settings, and I can't find them on Hubble images (0.04" per pixel...) either
Still see nothing better than or equal to NoiseXterminator.
NXT is really impressive for sure
I strongly disagree that Graxperts new noise removal is close in quality to NoiseXterminator. I've ran several tests and in my experience it's not even close. I'm using an ASI294MC Pro OSC camera, so I know it can very greatly depending on different setups. All that said, I still think its AWESOME that Graxpert is offering such a tool for free, and its a great way for people who don't have as much expendable income to get a great tool to start with for free. Kudos to Graxpert!!! I'm gonna go check out Topaz now...I own Topaz Denoise AI, but haven't used it in AP since I got BX....Lemme finish this video!
So yeah, I totally agree with drawing the line at adding details that aren't there, that being said, keep in mind that Topaz is NOT made for AP, it's made for everyday photography. Someone who is shooting a Hawk at 800mm that is distant may want those details drawn in for the feathers, for example. But us, as Astrophotographers, would never want artificial details added to our photos. Just imagine the field day the flat earthers would have with that?! So yeah, i'd probably mention that this is not a tool that is made solely for Astrophotography, rather, it is a tool made for photography. And for that, it does a pretty bang up job, all things considered.
Cuiv - Would you ever consider doing an interview on my podcast? We've grown QUITE a bit in the last few months, and i'd be honored to have you on to talk about AP one day. The episodes are on Spotify, Apple podcasts, pretty much everywhere. You can also just click on my profile as I upload every episode to TH-cam, but keep in mind I just started doing that a couple weeks ago, so don't judge by the view counts. We have 2500 listeners on all other platforms. I'd like to ask you about the stuff you don't get to talk about on your channel! Let me know, clear skies!
Roo
They are only tools. Use the right tool for the job. Topaz can be adjusted to eliminate these artifacts. It’s just a matter of mastering your tools.
I spent a LOT of time trying to make Photo AI not add those artifacts - the only way that I found also makes the denoising so weak as to be pointless...
How can you be totally sure that those details were not actually there 😂?
By comparing with Hubble images :)
@@CuivTheLazyGeek Ta réponse me convient et me convainc ;-)
👍👍 It's nice to have experience looking at these new tools since they seem to be coming out at an astonishing rate. It's almost like TH-cam... Click me! Click me!
Hahaha very true!