VideoHelp Forum

Our website is made possible by displaying online advertisements to our visitors. Consider supporting us by disable your adblocker or try DVDFab and copy, convert or make Blu-rays and DVDs! :)
+ Reply to Thread
Results 1 to 6 of 6
Thread
  1. I exported an AVS script containing the following commands as MagicYUV on dec. 30th :

    Code:
    LoadPlugin("C:\Logiciels autonomes\MeGUI\tools\lsmash\LSMASHSource.dll")
    LWLibavVideoSource("20131224_145353.m2ts", threads=1)
    
    FrameSurgeon(cmd="20131224_145353 FrameSurgeon.txt", show=false)
    
    Autoadjust(auto_gain=true, high_quality=true, gain_mode=1, input_tv=false)
    AutoExpose()
    HDRAGC(coef_gain=1.25, max_gain=3.00, coef_sat=1.0, max_sat=1.20, avg_lum=150, corrector=1.00, reducer=1.4, mode=2, protect=1, black_clip=0.50, shadows=true, debug=0)
    
    MCTemporalDenoise()
    
    ConvertToRGB(matrix="Rec709")
    Then I noticed that I had made a mistake in the FrameSurgeon frame interpolation template, made a few other corrections there, and re-exported the video as Lagarith (for some reason MagicYUV used to work fine and Lagarith gave me trouble, now it seems to be reversed, playback or export with MagicYUV sources produces a lot of black frames), with the exact same parameters otherwise. Then I took a PNG screenshot of a frame not located in the vicinity of the ones which had to be interpolated (not that many in that video, about 50 commands in total, some for several frames in a row, and I changed about 10 of them) : they look the same, but their sizes and checksums don't match (sizes are very close though : 1403371 / 1403321 bytes). I recompressed the MagicYUV file as Lagarith with VirtualDub2 in “fast recompress” mode, and took a screenshot of the exact same frame : the size and checksums do match those of the screenshot from the MagicYUV file, but of course not those of the screenshot of the newly exported Lagarith file, already identified as different.
    So is it expected in a case like this, to have the same filters with the same parameters processing the same source file produce a slightly different output ? Or is it the fact that a few frames were modified which had some sort of ripple effect in the processing of all frames coming after ? How could I identify in what ways those files are different ? I remember reading a thread, probably around 2015, where several (lossy) compression formats were compared with some kind of processing which outlined their differences only, compared with a reference file, but I can't find it again.
    Quote Quote  
  2. MTCD adds some randomness because of the gradfun dithering . It's like a noise generator, with a "random" seed .

    EDIT: just double checking it's only active with certain presets "low" and "very low" presets, so it's something else . But double(triple?) checking again, using show=true in the script demonstrates enhance=true so, it's active with default "medium" preset too, at least for the version I'm using. According to the little chart it should be false. But that's probably the reason. So the chart in the script and the reference url are wrong for "medium" and enhance. The others look correct. For v1.4.20
    http://avisynth.nl/index.php/MCTemporalDenoise
    Last edited by poisondeathray; 4th Jan 2019 at 22:41.
    Quote Quote  
  3. Alright then. But isn't “low” the default setting in MCTD ? And what is the purpose of adding random noise if the goal is to remove the existing noise ?

    Also, I noticed that processing a video with MCTD (default setting, supposedly “low”) increases the size of the output, compared with the same Avisynth script without the MCTD line, at least in Lagarith. Shouldn't it be the opposite ? (SMDegrain does reduce the size for instance, all else being equal.) Or is the benefit (with regards to file size / bitrate requirement) of reducing noise more than offset by the contra-sharpening treatment it applies by default ? (I remember I was told about this when I first tried this filter, quite a while ago – and after verification it was already you...)
    Quote Quote  
  4. Originally Posted by abolibibelot View Post
    Alright then. But isn't “low” the default setting in MCTD ? And what is the purpose of adding random noise if the goal is to remove the existing noise ?
    Default is low; use show=true . Enhance is enabled

    The purpose is dithering, to prevent ugly artifacts and banding

    Also, I noticed that processing a video with MCTD (default setting, supposedly “low”) increases the size of the output, compared with the same Avisynth script without the MCTD line, at least in Lagarith. Shouldn't it be the opposite ? (SMDegrain does reduce the size for instance, all else being equal.) Or is the benefit (with regards to file size / bitrate requirement) of reducing noise more than offset by the contra-sharpening treatment it applies by default ? (I remember I was told about this when I first tried this filter, quite a while ago – and after verification it was already you...)
    Yes dithering adds "noise" , but also the sharpening .

    You can adjust the parameter of any of them of course
    Quote Quote  
  5. In your experience, can it be made to compress better while still retaining its enhancing effect, or at least preserving a subjective level of sharpness similar to the source, or are these qualities mutually exclusive ?
    And does a lossy encoder, like x264, behave the same way with that kind of noise (i.e. increasing bitrate requirements at a given setting), or is it optimized to consider it as expendable information ?
    How does MCTD compare with other denoising filters, and have there been worthy new ones developped in the past few years ?
    Quote Quote  
  6. Originally Posted by abolibibelot View Post
    In your experience, can it be made to compress better while still retaining its enhancing effect, or at least preserving a subjective level of sharpness similar to the source, or are these qualities mutually exclusive ?
    Maybe a little bit if you fiddle with the settings

    A big problem , especially with 8bit is overdenoising . It predisposes you to banding / blocking artifacts . There are different types of dithering, you could use ordered dithering for example, that will reduce bitrate requirements , but is less effective generally at it's job

    It's a delicate balancing act


    And does a lossy encoder, like x264, behave the same way with that kind of noise (i.e. increasing bitrate requirements at a given setting), or is it optimized to consider it as expendable information ?
    How does MCTD compare with other denoising filters, and have there been worthy new ones developped in the past few years ?
    x264 requires more bitrate for for stronger dithering , random noise too. But you can tweak the settings to do a better job of retaining the dither and noise

    A overly simplistic way of looking at it, is the noise and dithering forces a lossy encoder to assign lower quantization values, thus higher quality. If you didn't do that , you end up with blocky artifacts and ugly gradients. Noise and dithering sort of "covers up" the defects , but requires more bitrate to maintain

    MTCD is still one of the best generic denoisers; but it also depends on the type of noise, characteristics of noise . Some may perform better or be more specific for certain situations
    Quote Quote  



Similar Threads