I've been working on something special...

Поділитися
Вставка
  • Опубліковано 27 січ 2025

КОМЕНТАРІ • 32

  • @deepskydetail
    @deepskydetail  10 місяців тому +6

    The thing that will make this model different than AstroSharp, is you will be able to tell the model PSFs for different parts of your image. If I can get it to work, it will take the PSFs and correct the star shapes. It's more than just sharpening, because the PSF you create will affect the output. Put in the wrong PSF, and you'll get wrong star shapes! Is it true deconvolution? Well, that's a topic for another video!

    • @salvadorpere
      @salvadorpere 10 місяців тому

      Thanks for your efforts to the amateur astronomy.
      We really appreciate what you are doing for us.
      Thanks so much

  • @jbmur
    @jbmur 10 місяців тому +1

    Thanks

  • @salvadorpere
    @salvadorpere 10 місяців тому +3

    Thanks for your efforts to the amateur astronomy.
    We really appreciate what you are doing for us.
    Thanks so much

    • @deepskydetail
      @deepskydetail  10 місяців тому

      Thank you for your kind words! I appreciate it!

  • @frankwakefield8815
    @frankwakefield8815 8 місяців тому +1

    I haven't used your tool yet but you are brilliant!!! Thank you for your work as it is amazing. I look forward to using it when I get a camera for my telescope.

    • @deepskydetail
      @deepskydetail  8 місяців тому

      Thank you! The model in this video isn't ready yet, but AstroSharp is still there :)

  • @jesuschrist2284
    @jesuschrist2284 10 місяців тому +2

    Astrosharp seemed down for me will give it another go. Ty for your work and videos

    • @deepskydetail
      @deepskydetail  10 місяців тому +1

      Thanks! Are you trying to use the website version of AstroSharp? That one is limited in size at the moment. The Github version (for desktop) is free though and should work. If it isn't working, send me an email, or join my discord and we can troubleshoot.

  • @materialvision
    @materialvision 10 місяців тому +1

    Thanks for your work, I have been working on a similar project using cyclegan for solar images lately and know how hard it can be, especially to make it useful for a wide range of input.

    • @deepskydetail
      @deepskydetail  10 місяців тому

      Ooo! That sounds pretty cool!
      I agree. It is pretty difficult to get all the parameters right. Training has failed on me dozens of times.

  • @carvrodrigo
    @carvrodrigo 10 місяців тому +1

    This will be a very nice tools for us. How much a RNN will mark features to correction in such hostile “image”? Just a curious question if this comes to your mind

  • @stevenickolls8016
    @stevenickolls8016 10 місяців тому +1

    Thank you for this. It goes quite over my head but you might want to see if the subject can be discussed further with Ivo Jager creator of StarTools.

    • @deepskydetail
      @deepskydetail  10 місяців тому +1

      Thanks! A lot of it goes over my head too, and it has taken me months to figure all this out!

  • @tsk1979
    @tsk1979 10 місяців тому

    what about images with newtonians with diffraction spikes. Even slight OOF or collimation can make the stars weird and then star reduction algorithms mess it up even further

    • @deepskydetail
      @deepskydetail  10 місяців тому

      Yeah, that's a hard one. The model can only deal with blur, tracking errors and coma.

  • @HyperX07
    @HyperX07 10 місяців тому

    Thanks for your efforts sir. I have a question that can astrosharp correct shapes of coma stars ?

    • @deepskydetail
      @deepskydetail  10 місяців тому

      The current version of AstroSharp has difficulty doing that. The tool I've been working on in this video should be able to help correct coma.

    • @HyperX07
      @HyperX07 10 місяців тому +1

      ​@@deepskydetail thank you sir. When the model will be published?

    • @deepskydetail
      @deepskydetail  10 місяців тому

      I am not sure yet

    • @HyperX07
      @HyperX07 10 місяців тому

      ​@@deepskydetailok. Thx in advance for the software

  • @whyf16uy
    @whyf16uy 10 місяців тому +1

    Man, you just continue to surprise me.

    • @deepskydetail
      @deepskydetail  10 місяців тому

      Well, don't be surprised yet! It might or might not work :)

    • @whyf16uy
      @whyf16uy 10 місяців тому +1

      @@deepskydetailI have great faith

  • @TevisC
    @TevisC 10 місяців тому +1

    I just bought blurXt :(
    Will your ai program be accelerated by GPU on the client? Big bad Nvidia seems to be the only supported guy in town : ( I have an Intel Arc A770 16gb. I think it can run Direct ML.
    I would certainly run your deconvolution tool if it ran on my GPU.

    • @deepskydetail
      @deepskydetail  10 місяців тому

      I think BlurX will still be better. Russell Croman is really good at what he does and understands this better than I do. But, I think this new thing I'm making is better than AstroSharp.
      Nvidia is widely used because they put out development kits more consistently than the others. It is my understanding that this let developers of AI programs (e.g., Keras, Torch etc) use them more easily. I'll try to make it multi platform. I'm not sure what's possible, but we'll see!

  • @Megawatt
    @Megawatt 10 місяців тому +1

    If this works, I will pay you for it.

  • @hacxman1
    @hacxman1 10 місяців тому +1

    "AI"

    • @deepskydetail
      @deepskydetail  10 місяців тому +1

      Yes. Marketing. I'm sorry! :)