VideoHelp Forum
+ Reply to Thread
Results 1 to 8 of 8
Thread
  1. Hello, I recently got a copy of the R2 version of Monster (2004) anime series DVDs. The quality is a lot better that anything that is out there, but could be better. They forced 4 chapters per DVD, and it shows. I have different issues with this project:

    Interlacing: I used QTGMC Interlaced (Slower). Speed is not a concern, but suggestions about anything more efficient, or that could result in better quality are welcome.

    Compression artifacts: I think these are typical mpeg2 artifacts, but I will provide samples for the experts to evaluate. Until now I tried these denoisers, with terrible results:
    • NLMeans (various setting, pretty bad)
    • TNLMeans (similar result to NLMeans)
    • TemporalDegrain2 (Maybe a tiny bit better than NLMeans)
    • MiniDeen (terrible, very smooth)

    Those are what I usually found recommended here for anime. So, I'm using StaxRip as gui, and this are my current filters:
    • QTGMC Interlaced (Slower)
    • Select Odd
    • TemporalDegrain2

    And the encoder setting, x264, rf10, very slow. I'm going for quality here, size and encoding time are secondary concerns.

    Video Samples:

    Image Samples

    - Here, towards the bottom of the picture, you can see the lines on the floor disappear, and I'm getting new artifacts, that are specially noticeable on the women's sweater.
    - In this one I'm also getting new artifacts, and the detail on the floor it's gone.
    As you can see, the offending parts, are also the ones that get the worst result. Suggestions are welcome, worst case scenario I'm gonna rip the chapters "uncompressed", but I hope someone can help me.
    Last edited by shark_bloody; 20th Mar 2021 at 22:49.
    Quote Quote  
  2. Don't use QTGMC on it. Before doing anything else you should IVTC it.

    And please provide an M2V cut from a VOB using DGIndex next time.
    Quote Quote  
  3. Originally Posted by manono View Post
    Don't use QTGMC on it. Before doing anything else you should IVTC it.

    And please provide an M2V cut from a VOB using DGIndex next time.
    Thanks, I will try with IVTC then, and I added a clip in m2v format.
    Quote Quote  
  4. Member
    Join Date
    Mar 2008
    Location
    United States
    Search Comp PM
    For the loss of detail in the flat parts look at aq-mode and aq-strength
    http://www.chaneru.com/Roku/HLS/X264_Settings.htm#aq-mode

    Here's a test I did with aq-mode=2, aq-strength=0.8

    This clip is reduced to 23.976 fps using IVTC, also a small amount of NR and some fine sharpening
    Image Attached Files
    Quote Quote  
  5. I did this:

    Code:
    Mpeg2Source("Sample1.demuxed.d2v", CPU2="ooooxx", Info=3) 
    TFM(d2v="Sample1.demuxed.d2v") 
    TDecimate() 
    Trim(3582, 4373)
    TTempSmooth(maxr=1, strength=2, lthresh=2, cthresh=3)
    gradfun3()
    And encoded with x264 preset slow, crf 16. Everything was done in 32 bit because the 64 bit version of Mpeg2Source() doesn't implement the deringing filters. This doesn't remove as much noise but retains more detail and has less posterization in the dark shots. The same segment as davexnet is more than twice as big though.
    Image Attached Files
    Quote Quote  
  6. Depending on the desired look, Anime4k might be interesting (no clue if it can be used in StaxRip).
    Attached a sample using VIVTV, Crop, Anime4k (resizing to 1024x770) at default settings just to give an impression of it's working.

    Cu Selur

    Ps.: I like the look of MiniDeen + CAS.
    Image Attached Files
    Last edited by Selur; 21st Mar 2021 at 06:26.
    users currently on my ignore list: deadrats, Stears555
    Quote Quote  
  7. You could try an open source upscaler like anime4k/waifu2x (or one of the forks), or commercial software like Topaz Video Enhance (I had very good results with it, especially with animated content) - the latter is rather expensive, but you can download... a trial version. It does wonders for compression artifacts.
    Quote Quote  
  8. I processed 30s of your original sample file with Topaz's deinterlacing and upscaling model. It does a slightly worse job than the non-deinterlacing models, which are older and I suppose trained with with more data. In any case, here is the result encoded in rf 10 h264. On the left is your original footage upscaled by ffmpeg to 1280x960, and on the right the result of Topaz processing.

    You can download it here.
    Last edited by starry_air; 22nd Mar 2021 at 11:27.
    Quote Quote  



Similar Threads

Visit our sponsor! Try DVDFab and backup Blu-rays!