Hello,
I have been using a temporal upscaling program (Dain App) to improve the smoothness of Jupiter rotation movies. The program uses artificial intelligence and machine learning to perform motion interpolation and thereby create extra frames which are inserted in order to increase the fps and smoothen the rotation movie.
The process involves, firstly, producing a rotation movie of a planet such as Jupiter. This is something that is relatively simple to do with several frames taken back-to-back. The frames are saved as a movie file. Usually, these movies appear jerky due to the discrete nature of each frame which usually results from a few minutes of video recording and stacking to produce a frame. However after temporal upscaling of the file, the result is a much smoother rotation movie. Below are links to a rotation movie (ORIGINAL) playing at 3 fps, and the same movie following temporal upscaling to 96 fps. Clearly, the temporally upscaled movie is more pleasing to view.
ORIGINAL rotation movie at 3 fps
TEMPORALLY UPSCALED rotation movie at 96 fps
I obtained the data over 2 hours of continuous imaging on October 10th 2021 from 9h 08m UT to 11h 11m UT with a 31 cm Newtonian, ASI224 camera and 3X TeleVue Barlow. 31 movies of 4 minutes each were first derotated using WinJupos followed by stacking with Autostakkert. Sharpening was achieved using Registax 6 wavelets followed by denoising and colour balance with Photoshop.
The ORIGINAL movie contains the original 31 frames. The TEMPORALLY UPSCALED movie contains about 32 times more frames.
The seeing at my site is typically poor (2-4 arcseconds), but the characteristic appearance of the dark poles on Io (and bright equator) are visible as it crosses the face of Jupiter. An artefact of the upscaling process is visible when Io crosses the limb of Jupiter, something I am trying to find a solution to. The temporally upscaled movie shows no jerkiness and appears much more natural.
Obviously, since the program is creating new frames, a good question is whether these new frames could be used to create a better image through derotation and stacking of the images. Since the images created by the AI have similar information to the original frames, but the noise may be stochastic and will therefore be averaged out, it is possible that the S/N might be improved. If you think this sounds like black magic, then that's AI. But I'm not sure if it will actually work.
Several AI temporal upscaling programs are available. I used
DAIN app since it is relatively inexpensive, however it only works with NVIDIA graphics cards. I am not affiliated with any of the programs I have mentioned here.
Zac Pujic
