BlurXTerminator - AI sharpening for astrophotography
Вставка
- Опубліковано 16 гру 2022
- You can find BlurXterminator here: www.rc-astro.com/resources/Bl...
Blurx is an AI based deconvolution program which runs as a pixinsight plugin. In this video I show how to use the program, compare it to the old way of deconvolution, and talk about why I think this tool is really worth it. The truly revolutionary part about this program is to handle images where the PSF is not coherent globally, i.e field rotation, tip tilt, coma, etc. This by itself makes the program worth it.
#astrophotography #telescope #andromedagalaxy #pixinsight #zwo #orionnebula #milkyway #astrofalls #blurxterminator - Наука та технологія
What a great explanation on deconvolution. I really learn something today! Thanks
“Revolutionary” is the word that comes to mind. I just bought this last night and wow.
Excellent review and explanation Bray! I had used the BXT plug-in. Its result is amazing and more importantly it made decon so easy and effective. I had gone down the Dynamic PSF picking stars and the regular deconvolution task route before. It was quite a pain…BXT made that step insanely easy and yields excellent result. That alone made it a good purchase. The “global” examples are quite interesting and convincing.
Thanks glad I could help! The time saving factor is one of the best parts
That video is absolutely PACKED with knowledge and information. Insane.
This looks like a great tool! Thank you for such a clear explanation.
Really nice demonstration/explanation Bray.
Thank you!
Great review and images - on my shopping list!
Excelent explanation Bray!
Thanks!
A good oversight. Concise and to the point as well!
Thanks Neil!
Just landed in your channel. Very nice explanation of the Devon process.
Subscribed 😊.
Thanks! Glad you liked it
Great summary!❤
Great review and tut Bray!
Thanks for the video, BlurXterminator also benefits from CUDA like the other Croman processes. He needs to get the GradientXterminator in PI too. Clear Skies and Merry Christmas!
Full beans, I'm gonna use that now.
Interesting Video trank you
Hey Bray, have you tried adjusting your backfocus to address your corners? It looks like your sensor is too close to your flattener.
Hi Bray
Excellent video as always.
At what stage do you recommend doing the Deconvolution process with BlurXterminator?
During the linear stage after DBE or after stretching when the image is in non linear stage?
Thanks!
It is best to apply in its linear state before doing anything else, just like normal deconvolution. You could do some cropping before hand
Hi and thks for the good video. I feel a bit frustrated after applying NXT on different linear astrophotos, I always get lots of extra noise and weird artifacts. I've tried using the default parameters and I've also tried different parameters, but the result is always the same. I mean, I see the stars on my images get pinpoint ones, and in some cases, some areas of my nebula photos get a bit better (more details and structure), but the noise is more than the original photo. I have only applied ABE or DBE to my images, with even no color correction, but I cannot get ridículo of that noise. Could you help me please? Any suggestions/ideas? Parameters to try? Thks in advance
How do you think it works for wide angle milkyway shots?
O tradutor ajuda muito, fico feliz que seu canal o tradutor funcione mas não traduz bem.
No caso uma imagem foi registrada por um telescópio de certo modelo, essa imagem meio desfocada com baixa nitizes...
Esse programa adaptou e otimizou a imagem ou reduziu naturalmente a ampliação, assim melhorando a imagem?
Não sei se vai responder, recomendo acompanhar o maior canal de divulgação científica do Brasil e talvez do mundo..
. Space today ❤️
Boa madrugada
It would be interesting to perform the following experiment: Run BlurXterminator on each frame of an image stack (sizable stack). Then stack adjacent pairs of BlurXterminator frames, and run BlurXterminator on the resultant image frames. Repeat until only one image is returned at the end. Then compare this image with the BlurXterminator image of the fully (raw) frame stack.
While this is a fair amount of work, if it results in superior image, then this procedure can be automated.
You could do this with an image container but would be a bit overkill IMO. The neural net will also work much better on a stacked image with higher SNR
Estes dias eu tava pensando, na possibilidade de um algoritmo ou software, para adaptar ou fazer tipo interferômetria óptica. Só que com duas imagens captadas por um mesmo telescópio, só que com distância de câmeras diferente, o algoritmo apenas pegaria a imagem final obtida pelos ccds oi CMOS, e como interferômetria... Somar as imagens e assim maior nitizes e resolução. Ideia sem noção, mas foi uma coisa que veio a minha cabeça.
Hi , you miss Select AI and I do not see any File under name AI :(
I'm still very new to astrophotography, so a lot of what you're discussing here is still foreign to me. That being said, if I'm understanding your explanation correctly of BlurXterminator and how it uses existing images from other telescopes such as Hubble to build out additional details in your own images, at what point does your image become NOT your image? Forgive me if I'm misunderstanding though, as I'm not hating on the program just wanting to learn.
This is a very good question! The answer to this is pretty indeterminate and depends on how the AI model is deployed. There is a point where you could run the program so aggressively that you do approach that “no longer your image” point, and you always want to avoid this. However the same question could be asked of the traditional deconvolution method, since you are really guessing what the PSF should be. The answer to when it is too much is down to taste, and you will get an eye for it over time
@@astrofalls Thank you!
This is not how neural networks work. HST and JWST images are used during training only because they have very sharp resolution with typical astronomical structures. The network learns these structures only in a "probability distribution" sense, not in a memorization sense. It can't and doesn't "paste in" detail from images it has seen during training - it learns to perform deconvolution within the context of astronomical images.
Deixo os comentários com meu devaneio final...
No processo de interferômetria óptica usada pelo Very large telescolpe no Chile, eu imagino dois telescópios do mesmo tipo, conectados para interferômetria, assim ter uma nitidez x a distância dividida pelo diâmetro do espelho do telescópio. Futuramente eu quero fazer isso, por isso trabalho muito para conseguir dinheiro. Dinheiro é sempre bom problema, não os equipamentos em si
Comparable to topaz ai sharpen?
Not even a worthy comparison IMO. Topaz is useless for DSO astro
Topaz AI is good if you want to grow fur or feathers on your galaxy images :D