Personal View site logo
Windmotion - upsampling to 4:4:4
  • 263 Replies sorted by
  • @icp the likely culprit for your loss of detail is, I'm betting, the avisynth solutions do their thing to not just the color information. The luma channel is fully sampled and will have very little to no noise compared to the blue and (sometimes) red channel. You can get the benefit of the noise reduction and restore your lost detail via something like After Effects. Layer your de-noised footage on top of the same clip in un-altered form and change the transfer mode of the de-noised footage to COLOR.

  • @rean Thank you for taking the time to develop this. I was curious if you or other members have suggestions to use Windmotion on an OS X system? Thanks.

  • I've keep my eyes on this project, I'm always looking a way to improve the footage I get from the camera.

    I've used avisynth in the past a lot. It's incredible flexible and powerful, but not for newbies, you have to know (somehow) what you're doing. For denoising: Neat video beat anything I've used in avisynth, in avisynth I've always suffered for lost of details that NV don't. For film grain, I've used grainfactory a lot, it's amazing (and slow) script, but in the past years I've used match grain in AE.

  • @rean

    It is impossible to use it "inside" and also as a plugin due Avisynth architecture limitations. Also different stages (~20) are used with a special visual control for every stage. Сurrently the program is for prepoduction only.

    The alternative would be to build it as an installable codec that would replace the bundled MainConcept H.264 codec in both After Effects and Premiere Pro. This is how GoPro's CineForm codec is integrated into CS5 and CS6, using either AVI or Quicktime as file containers.

  • @jazzroy not all strings were applied. Install any official avisynth version or wait for Alpha3.

  • @rean the strings for avisynth location weren't written, so I added them, but the .bat test file continues reporting same error.

  • @zsero Sorry, but I cannot implement Cineform avi export because I know no way to pass 16 bit/channel video from Avisynth to DirectShow. Also ffmpeg/ffmbc have limited support of 10-bit codecs. Currently only DNxHD 4:2:2 10 bit and H.264 10 bit (all chroma sampling) can be used to export 16-bit video. ProRes support is very strange. I get 8-bit video even from a 16-bit source. So, probably I will code my own c++ code to implement export later, but not in a near future.

    A question for GH2/3 owners. What a minimal size in bytes of a first 4GiB MTS file (spanning)? This data is required to implement correct import from your camera.

  • @jazzroy check reg file values in the registry. I had similar problems on Windows 8 x64. This is a next "super-puper" microsoft feature to not store reg files located not in Program Files. The next release will use installer to store values to registry, so we will have no similar problems.

    @zsero ok. Will try to make different files in the future. But, anyway, the difference in very visible in the program.

  • @rean, I think it's not easy to see a difference, because you are comparing them side by side. It's almost impossible for a human visual system to compare things side-by-side.

    A better way would be just links to the original 2 source images and then you can just change tabs in your browser. Also, there is a special website just for comparing two images by mouse hover, but I forgot it's name.

  • Hi Rean, I'm using a Nex-5N and I'm reallyinterested in trying your software, but I get a "Avi import filter error" when I do the 04.Check Avisynth and VirtualDub.bat.

    Followed the steps carefully, placed Soft in C, run the .reg, installed the three programs.

    Running Windows 8 64 bit.

  • Oops... imgurl recompress the png source :(

    Reposted again... PS. Please see in the real 100% resolution. Web browsers scale the source to fit it to a window.

  • Maybe the jpeg is masking the difference.

    I have posted a 24-bit png from a 16-bit 4:4:4 original.

    Of course, some users cannot see any difference. Most of users cannot see a difference between Youtube and a HQ video in 4K. Most of them use low quality systems, like a mp3 player for $1 and very happy with it.

    Windmotion is not for these users.

  • I guess if you cant see the difference in the little girls before and after comparison then you probably don't need this software. Anyone else not seeing a difference in the bottom still? I can see the blocking in the top one that is fixed in the second one, but no difference in the side by side. Maybe the jpeg is masking the difference.

  • @Burnet:

    Is there an equivalent to Quicktime reference movies in any AVI player?

    It's called frame serving, and there is more than one product for it. My favourite is the Debugmode Frameserver: http://www.debugmode.com/frameserver/

  • Heh, some of it might not be legal!

    Oops, sorry. I have moved it to another site. zsero, thanks for the recommendation.

    Is there any chances to get your code inside a After Effects plugin?

    No. It is impossible to use it "inside" and also as a plugin due Avisynth architecture limitations. Also different stages (~20) are used with a special visual control for every stage. Сurrently the program is for prepoduction only.

    Probably you can export your project files, then import it into Windmotion, process it there and get it back again. What a file format After Effect uses for project files? Probably I can create import and export for it.

    Warning. It is not a correct way to use Windmotion, but you can try it for your old projects. The new projects should use this scheme:

    1. Get a *.mts source.
    2. Import it into Windmotion.
    3. Proccess it in a batch mode.
    4. Export it to a file format you need for your software.
    5. Your own tasks.
    6. Remove any temporary files.
    7. Keep original *.mts source and Windmotion project files (~3 kb per a source) to a backup disk array or a tape.

    Because Windmotion uses text files for projects, there is no needs to keep temporary and end results. You can restore them later in any time.

    To restore a project in a future:

    1. Restore all *.mts source and Windmotion project files.
    2. Repeat all previous steps (just mouse clicks and waits for the result. No any editing!)

    Also, if you have a mistake in any stage, you can try to correct it later. There is no any "source destroy" stage, and a need to repeat all the steps again like in editors.

    Windmotion has a command principle, used in a node-based video software. I have no reason to use a editing software like Adobe or plugins, because I cannot change any middle stage later. Using “editors” is not a correct way for video production now. All the future video software will be command- or node- based.

    Another question would be how does the Avisynth noise reduction compare to Neat Video? In case Neat Video is better, could you chain it in between at the right place?

    Avisynth has no noise reduction. It is a platform for other plugins. Currently I use different plugins and special algorithms to use them specially for the program tasks. Also Windmotion has a possibility to choose a better way to denoise in every special case. It is very creative stage. It is not a simple solution "click and get a result".

    So it is a Neat Video alternative solution. You can try it to get a compare. For me personally, I really have no reason to use Neat Video, because Windmotion suits me for any noised sources I have.

  • How is being Quicktime free an advantage in Windows? Is there an equivalent to Quicktime reference movies in any AVI player? That alone would prevent me from ever being Quicktime free in Windows.

  • @Meierhans,

    If he'd integrate it into AE, it'd all be limited to whatever decoder AE uses. For example look at the "digital rain" problems, the Adobe decoder produces this bug. Maybe Adobe will fix it, maybe not. The only workaround is not to use Adobe as a decoder, but something like 5DtoRGB.

    So this tool is an alternative to 5DtoRGB / Cineform HD Link, not a plugin. It can use whatever decoder it needs, if we find one to be buggy, @rean can just change the decoder. Not with Adobe.

    Also, I think it's perfectly fine to import a footage into an editing friendly format (what is not h264). In my opinion the best is Cineform, since it's free and it's AVI based, unlike ProRes and DNxHD, both what are MOV based. So you can have a totally QuickTime free workflow on windows (a big big advantage!).

  • WOW! Is there any chances to get your code inside a After Effects plugin? Your results look promising, but the fact that you need to apply the "effect" before your start grading means one would either need to convert all his raw footage, or do another round trip after edit lock, convert all used clips, replace... Both is not really nice way to work. Another question would be how does the Avisynth noise reduction compare to Neat Video? In case Neat Video is better, could you chain it in between at the right place?

  • Yes, that's quite a disgusting site, @rean, I'd recommend you to use imgur.

    A very handy open source program to upload all kind of things to image sharing websites (as well as for lot of other ones as well) is ShareX: https://code.google.com/p/sharex/

  • ... the advertising on the site you linked those images may not be work safe.

    Heh, some of it might not be legal!

  • @rean, might want to put some forewarning that the advertising on the site you linked those images may not be work safe.

  • Some results from Sony FS100 ($6,550.00 U.S. List Price) on PreAlpha3 (version in development):

    Chroma channel artifacts (before, after and difference):

    links are removed

    Here the processing includes: H.264 predication errors fixing (28mbps), black level correction, small color correction, non-halo-producing micro-contrast sharp, denoise, dithering, conversion from 4:2:0 to 4:4:4.

  • Watching this topic. sounds interesting.

  • @spacewig No, I have no compare. I have no GH2. Restoring of an "original" image is not possible, so results will looking different in compare. Unable to get information where it is not available (using a lossy codec). However, the image can be adjusted closer to the feel of the feelings from the real image. There are a variety of ways. Widely known is, for example, Dithering, to masking low-level digital artifacts. Missing bits of information in the program are obtained by interpolation using motion prediction and interpolation using the noise reduction. Aliasing is removed with the use of algorithms based on neural network training. And so on.

    Since all this is very difficult to develop alone, I use the GPL code, already implemented on a lot of Avisynth plugins. Therefore, the program has its own specific limitations, related to limitation of all these components. I added the methodology for applying the algorithms, using my personal experience to handle video and photos.

    About sharing. Thank you. I will contact you in PM.

  • @rean Have you compared the results of restoring hacked with unhackd GH2 footage?

    I'd be happy to share original MTS files that you can then publish. Just let me know how to get it to you.