Why I'm using DeepSNR For Noise Removal and You Can Too

good day Dylan here from the byon bay Observatory I've just finished a photo uh which is a miracle because we've got some clear weather so check out this image of m58 the m58 M83 the southern pin wheel Galaxy this is a beautiful classic looking Bard spiral galaxy one of the best in the southern hemisphere I think for these normal sort of looking galaxies because we've got plenty of cooky looking galaxies but this one is a classic in this image I did something differently in my processing than I've ever done before now I've come at Den noising images a number of ways over the years but recently I've fallen into basically using a little little tiny tiny tiny little dab of topaz AI D noising at the end of my workflow it makes everything so buttery smooth it looks amazing but completely unnatural so what I end up doing is layering those and then pulling back the blend Just a Touch so I might blend in 50% topaz Doo with the original image so the original grain comes through just a little bit so it doesn't look too unnatural but there's a problem with this topaz's AI model is trained on terrestrial images it knows nothing about astrophotography it's suspicious in that it may especially with the sharpening it may introduce artifacts or details that aren't really there and that is always something we want to avoid in AST Photography in general so we do have to be careful with using AI tools so when tools come along like blur X which are trained on astrophotography data trained on real data that's something I would prefer to use over the more General Solutions like topaz Ai and I have an exciting new plugin for PX Insight that I used on this M83 image and it kind of blew me away I'm going to be talking about deep SNR this is a plugin from the same guy who coded starnet which we use a lot for Star removal but I want to show you something I discovered that applies to starnet and blur X and this NE toal deep SNR and I'm going to show you how I can get the best possible D noise result out of my images now let me show you how it works my name is Dylan odonnell you're watching Star stuff [Music] you might have noticed that not every video has a sponsor I only put a sponsor on videos which I think are really useful because I know that they'll get more views that makes the sponsor happy too so I've got no hesitations because this is going to be a really useful video for you guys to make sure that I shout out High Point scientific High Point scientific are a New Jersey astronomy vendor they sell telescopes cameras everything you need for your astrophotography journey they fully support their product and they have a price match guarantee and I believe they just opened up international shipping they've been supporting this channel for a long time and I really enjoy that because they get great feedback from my viewers and from their customers so check out www.h highpointscientific all the links Below in the description if you want to replicate my setup all the links to all my gears down there okay let's use deep SNR now okay so funny story I actually reached out to Topaz fed back to them and I said can you train your tool a little bit more on astrophotography because it doesn't seem to deal with Stella stuff very well and they said yeah that sounds like a great idea I haven't heard a word from them since uh so this tool actually is something I've been waiting for for a long time I really love what the author has done here now this is the website deep SNR astro.com and the author is Makita Mur sounds like a full-on rockstar name like a guy who invented a cryptocurrency and then released the white paper and then disappeared uh but he has a fantastic Gallery if you haven't checked it out uh he is severely under rated 247 followers amazing anyway I shouldn't have to tell you how to install software and I'm not going to go through the process in detail but essentially just go to the download page on that website download the um version for Mac or Windows as you need and you'll have these files in a zip file in your downloads folder just drag and drop them into the bin folder in your pix Insight this C program files pick Insight bin just drag and drop them in then when when you open pick Insight you just have to install that component we go to processes modules install modules search it finds it there's the Deep SNR and we just say install and then you should be ready to go pretty easy now let's open up a test image okay so here is the final image before I apply any of the D noise and Sharpening at the end of what I do so this has been taken with three filters uh G sorry ha A G and B for the Galaxy combination and because I've drizzled the image and I'm taking it such a huge resolution the pixels on the qy 268m are quite small I end up with an image usually at 4,996 pixels which is quite large uh so I typically downscale from here and then get good sampling with my setup but let's look at this noise profile I'll zoom in even more because I want this to come through on the recording I'll Zoom right into these little peripheral galaxies on the edge here okay so we have some noise it's not too overwhelming but it's something I'd like to reduce in the final image now if I run this deep SNR process let's have a look there's not much to the tool right it's got one slider for strength and a tick for linear data now this is linear data because I have already run the stretch and applied the stretch with histogram transformation in pix Insight oh and I should mention also this plugin is free I didn't have to pay for this but P inside is not free so you do have to bring your own P inside all right so I've got my nonlinear data we're told by the author and by people who have used it that it doesn't work well with debed images it only works well if you're using if you're using astrophotography which has been shot on a monoc chrome camera and then Channel combined later because the type of noise that one shot color debing introduces is not the type of noise it's trained on to remove but let's see how this goes this is going to take a while I'm going to drag and drop from the triangle onto the image and it goes through its process and let's let the magic of editing speed this up a bit cuz it takes a while go so you'll see I'm not really happy with this result it still has a lot of speckle all through the image and it's an odd kind of speckle I'm just going to clone this out as a cop and see where we're at you can see all that speckle there it's not great if this was what deep SNR does I wouldn't be recommending it but let's leave this here so we can compare because I've found okay so there's the before and after you can see there's a lot of red color noise in here and here it's reduced a lot of that but it still has this uh Red Dot speckle all the way through it but here's the thing I have found that these AI tools work wildly differently at different scales and the author of this tool has said that you should be doing this early on in your workflow doesn't apply to me at least because images are huge in fact this image was drizzled and it was much bigger than this even and I've pulled it down to 4096 but I find if I pull the image down further to get it to web size or screen size the way I want it not print size but screen size then run the tool again completely different results I'll show you what I mean so now I'm going to resample this image I'm going to resample it by half to 2048 now I'm going to run deep SNR again and let's see how it's different that was a lot faster a lot faster than topaz AI but you see now I have this result which is buttery smooth like topaz but pretty clean like it's preserved all the stars and the shapes really well that is to me perfect noise reduction I might even back it off a touch but otherwise that's great and I find that trick with going down to a manageable size or a display size actually works really well for other AI tools as well like blur X I'll just run this on defaults here and see how it goes see blur X has this reputation for looking overcooked but when I come down to this size it's still a little bit crunchy but it's not as overcooked as it looks at other sizes so I can do less blending I could get away with posting this image as a complete image right now I did back it off a little bit in my final so I back it off with blending in Photoshop layers just so I can have more granular control over the output but I am really happy with this image and I'm really happy that these tools are available for us to just take that last little step the final step in the processing process the processing process so deep SNR I highly recommend it but just tell me what your results are with this science trick if you have an image which is 10,000 pixels wide or 4,000 5,000 pixels wide I find that these tools aren't doing a very good job but at screen resolutions they are absolutely fantastic so it might be harder to use these tools for print if you're doing high high resolution stuff that's intended for magazines or posters or Billboards or whatever um but certainly for screen sizes these tools are excellent I find that deep SNR gives me a result that I like better than topaz and so from now on I will be switching to deep SNR as my default D noise tool it's a pretty shock before and after really isn't it uh thank you for watching this video I hope this little tidbit helps your processing workflow if you don't use pix Insight you can't use deep SNR yet however the author has been known to release Standalone versions uh also this isn't being GPU accelerator just yet but starnet does have GPU acceleration support so there's every possibility in future that this D noise tool might also get graphics card acceleration which is fantastic thank you for all your great great work Nikita I highly recommend everyone jumps over and follows him on Instagram and also tries out this plugin I highly recommend it that's it my name is Dylan oald and you've been watching Star stuff and remember everything is meaningless and we're all going to die [Music] he

Channel: Dylan O'Donnell

Views: 8,397

Keywords:

Id: FJHFIWOo4IQ

Length: 11min 10sec (670 seconds)

Published: Thu Jun 06 2024