VideoHelp Forum

Our website is made possible by displaying online advertisements to our visitors. Consider supporting us by disable your adblocker or Try ConvertXtoDVD and convert all your movies to DVD. Free trial ! :)
+ Reply to Thread
Results 1 to 21 of 21
Thread
  1. there is this guy i know nothing about

    all he does is upload anime series he upscales to torrent sites
    he never comments for some reason (maybe he doesnt understand english, he is Chinese so maybe he hangs at other chinese websites ) (if some one here knows chinese and could help me dig something on him ill be glad , i have a bit more information on them\him)

    he doesn't rip the DVD himself
    he just uses the 480p rips other people upload

    here are some examples (one i a regular 480p rip the second is his upscale 720p)

    http://screenshotcomparison.com/comparison/192192

    http://screenshotcomparison.com/comparison/192194

    also he releases them pretty fast...

    how does he make the lines look so much better?

    i tried every thing with sharp() and aWarpsharp() its not just this...

    the reason i am asking is because i want to know how he did it and use it on some other shows i have.
    Quote Quote  
  2. He's probably using a dehalo_alpha(), a line darkener (fastlinedarken, fastlinedarkenmod), and an antialiaser (santiag, aaa) too. And upscaling with nnedi3.
    Quote Quote  
  3. Mod Neophyte Super Moderator redwudz's Avatar
    Join Date
    Sep 2002
    Location
    USA
    Search Comp PM
    zanzar, in the future please use a more descriptive subject title in your posts to allow others to search for similar topics. I will change yours this time. From our rules:
    Try to choose a subject that describes your topic.
    Please do not use topic subjects like Help me!!! or Problems.
    Thanks,

    Moderator redwudz
    Quote Quote  
  4. I suspect he's using Hysteria() too. That gives line darkening and thickening.
    Quote Quote  
  5. Originally Posted by jagabo View Post
    He's probably using a dehalo_alpha(), a line darkener (fastlinedarken, fastlinedarkenmod), and an antialiaser (santiag, aaa) too. And upscaling with nnedi3.
    SetMtMode(5,4)
    v = LWLibavVideoSource("d:\4.mp4")
    a = LWLibavAudioSource("d:\4.mp4")
    AudioDub(a,v)
    ConvertToYV12()
    ColorMatrix(mode="rec.601->rec.709")
    fastlinedarkenmod()
    santiag()
    nnedi3_rpow2(2, cshift="spline36resize", fwidth=1280, fheight=720)

    and it looks like dehalo alpha blurs all the barely seen lines and makes it worse (is there a way to restore them)

    this is the end result http://i.imgur.com/kqdHGdM.jpg

    far from his http://i.imgur.com/v6t4f55.jpg

    is this really this hard to imitate this? i think my knowledge on this subject is just limited

    here is another exmaple of other series
    http://screenshotcomparison.com/comparison/192224 i know he uses other people rips because some times a problem that appear in some ones rip also appear in his version (defects that cannot be removed with filters)

    it looks like there is no mistakes there just a better quality and thats it -perfect-

    is it really that hard to archive it? even with your knowledge? you basically answer every question on this sub with ease
    Last edited by zanzar; 29th Nov 2016 at 12:44.
    Quote Quote  
  6. Did you try using Hysteria()? Can you upload a sample of your source?
    Quote Quote  
  7. I would give a chance to https://en.wikipedia.org/wiki/Dilation_%28morphology%29 first (prefer to oversample input video and after this apply dilate kernel) - it will be good to remove noise and distortions/artifacts from signal - perhaps with help of http://www.compression.ru/video/cartoon_restore/index_en.htm or similar filter.
    Quote Quote  
  8. Originally Posted by jagabo View Post
    Did you try using Hysteria()? Can you upload a sample of your source?
    Originally Posted by pandy View Post
    I would give a chance to https://en.wikipedia.org/wiki/Dilation_%28morphology%29 first (prefer to oversample input video and after this apply dilate kernel) - it will be good to remove noise and distortions/artifacts from signal - perhaps with help of http://www.compression.ru/video/cartoon_restore/index_en.htm or similar filter.


    sure

    i created a rar with 3 different samples from different shows one 480p raw and the other is his 720p version

    you can actually see the "filtering style" in all of them, "buttery" clean lines , no noise and it looks very nice to the eye (its like he uses the same script for all of them) compared to the original.
    the general improvement in quality varies from show to show

    thanks for trying to help

    fixed
    Image Attached Files
    Last edited by zanzar; 29th Nov 2016 at 17:15.
    Quote Quote  
  9. Originally Posted by jagabo View Post
    Did you try using Hysteria()? Can you upload a sample of your source?
    fixed the samples to be good
    Quote Quote  
  10. Originally Posted by pandy View Post
    I would give a chance to https://en.wikipedia.org/wiki/Dilation_%28morphology%29 first (prefer to oversample input video and after this apply dilate kernel) - it will be good to remove noise and distortions/artifacts from signal - perhaps with help of http://www.compression.ru/video/cartoon_restore/index_en.htm or similar filter.
    after reading it it sound like a good idea but how do i actually use it?
    the first link is just a wiki and the second looks like something not related to avisyth

    sorry if it makes me sound , sorry for the lack of understanding
    Quote Quote  
  11. Originally Posted by zanzar View Post
    after reading it it sound like a good idea but how do i actually use it?
    the first link is just a wiki and the second looks like something not related to avisyth

    sorry if it makes me sound , sorry for the lack of understanding
    No need to sorry - it was ad hoc idea - you can give a chance to ffmpeg (it is quite fast) bellow is example for ffplay to perform this kind of processing while watching video.

    Code:
    @set vproc="pp=ac/dr,hqdn3d=2,scale=iw:ih:sws_flags=neighbor:sws_dither=none:in_range=1:out_range=1,xbr=2,erosion=threshold0=65535:threshold1=0:threshold2=0:threshold3=0,format=pix_fmts=yuv420p,zscale=d=ordered:f=spline36:r=full:w=iw/2:h=ih/2"
    @ffplay.exe -hide_banner -color_range 2 -loop 0 -i "%1" -an -sn -vf %vproc%
    Second link was related to VirtualDub (where i would recommend modern fork https://sourceforge.net/p/vdfiltermod/wiki/Home/ ) but Avisynth is able to use Virtualdub filters, ffmpeg is able to use avisynth scripts too - there is many combinations you can use - sometimes ffmpeg or virtualdub can be faster/better sometimes avisynth can't be replaced...

    And i need to apologize - i wrongly named erode filter as dilate - so erode is filter you searching for.
    Last edited by pandy; 1st Dec 2016 at 04:47.
    Quote Quote  
  12. Originally Posted by jagabo View Post
    Did you try using Hysteria()? Can you upload a sample of your source?
    Hysteria seems to be working well (better then other sharpeners) but i still didnt get close

    SetMtMode(5,4)
    v = LWLibavVideoSource("d:\4.mp4")
    a = LWLibavAudioSource("d:\4.mp4")
    AudioDub(a,v)
    ConvertToYV12()
    ColorMatrix(mode="rec.601->rec.709")
    Hysteria(strength = 2.0 )
    DeHalo_Alpha(rx=2 , ry=2)
    SeeSaw()
    TemporalDegrain(SAD1=600, SAD2=400, Sigma=18) *doesnt seem to help denoising very efficiently...
    nnedi3_rpow2(2, cshift="spline64resize", fwidth=1280, fheight=720)

    Does placing dehalo after hysteria counters the faded line distruction? Since it darkens them they are less affected right?

    result http://i.imgur.com/EzEomqA.jpg

    his http://i.imgur.com/duBEz7K.jpg


    it looks close only from afar if you zoom in just a bit there load of difference (one is 720p and the other is 1080p is because one is a screen shot and the other are from avspmod)

    his version has much smoother result (no noise or artifacts and softer and more exact lines...)

    were you able to scramble something up ?
    Image Attached Files
    • File Type: mp4 4.mp4 (2.75 MB, 114 views)
    Last edited by zanzar; 30th Nov 2016 at 18:10.
    Quote Quote  
  13. waifu2x does a pretty good job for upscaling anime, but the avisynth version is too slow to be useful (vapoursynth version is about 20-30x faster, but still only get <1fps)

    Something is messed up with the timestamps in your sample. Why don't you just cut a piece of the original source ? Why are you using VLC ?
    Quote Quote  
  14. e.g. waifu2x "default" anime settings, no other filters except colormatrix
    Image Attached Files
    Quote Quote  
  15. Hysteria has lots of parameters you can play around with. And you can call it more than once. Something like this is getting a little closer:

    Code:
    LWLibavVideoSource("4.mp4") 
    ColorMatrix(mode="rec.601->rec.709")
    Hysteria(strength=5.0, maxchg=20, lowthresh=15, highthresh=20)
    Santiag(1,1)
    McTemporalDenoise(settings="high")
    nnedi3_rpow2(2, cshift="spline64resize", fwidth=1280, fheight=720)
    Hysteria(strength=3.0, maxchg=20)
    Santiag(3,3)
    Dehalo_alpha(rx=3, ry=3, darkstr=0.5, brightstr=0.5)
    nnedi3_rpow2(2, cshift="spline64resize", fwidth=1920, fheight=1080)
    Blur(1.0)
    It's creating some dark edges where it shouldn't though. Like in the shadow beneath the sunglasses. That should give you some ideas.

    Are you aware that you can use Interleave() to view changes caused by a filter? For example:

    Code:
    Interleave(last, Santiag(3,3))
    will let you easily switch back and forth between the video without Santiag() and the video with it. That, along with a screen magnifier, makes it easy to see exactly what a filter is doing.
    Quote Quote  
  16. Originally Posted by jagabo View Post
    Hysteria has lots of parameters you can play around with. And you can call it more than once. Something like this is getting a little closer:

    Code:
    LWLibavVideoSource("4.mp4") 
    ColorMatrix(mode="rec.601->rec.709")
    Hysteria(strength=5.0, maxchg=20, lowthresh=15, highthresh=20)
    Santiag(1,1)
    McTemporalDenoise(settings="high")
    nnedi3_rpow2(2, cshift="spline64resize", fwidth=1280, fheight=720)
    Hysteria(strength=3.0, maxchg=20)
    Santiag(3,3)
    Dehalo_alpha(rx=3, ry=3, darkstr=0.5, brightstr=0.5)
    nnedi3_rpow2(2, cshift="spline64resize", fwidth=1920, fheight=1080)
    Blur(1.0)
    It's creating some dark edges where it shouldn't though. Like in the shadow beneath the sunglasses. That should give you some ideas.

    Are you aware that you can use Interleave() to view changes caused by a filter? For example:

    Code:
    Interleave(last, Santiag(3,3))
    will let you easily switch back and forth between the video without Santiag() and the video with it. That, along with a screen magnifier, makes it easy to see exactly what a filter is doing.
    well thanks i think it works i played with the script till i've got almost the same result

    a = LWLibavAudioSource("d:\4.mp4")
    AudioDub(a,v)
    ConvertToYV12()
    ColorMatrix(mode="rec.601->rec.709")
    ColorYUV(cont_y=0, cont_u=0, cont_v=0,gamma_y=0)
    Hysteria(strength=5.0, maxchg=20, lowthresh=15, highthresh=20)
    McTemporalDenoise(settings="high")
    nnedi3_rpow2(2, cshift="spline36resize", fwidth=1280, fheight=720)
    Hysteria(strength=4.0, maxchg=20)
    Dehalo_alpha(rx=3, ry=3, darkstr=0.3, brightstr=0.3)
    nnedi3_rpow2(2, cshift="spline36resize", fwidth=1920, fheight=1080)
    Blur(1.5)
    aWarpSharp(depth=14)

    i found that "Santiag" worsen the quality (dont know how to describe it with terms but here is a link where the first is without it and the second with)

    http://screenshotcomparison.com/comparison/192430

    of course its not the same sample you used but it looks like here specifically it doesnt help .... do you agree?
    how can i know when to use "Santiag" and when not? and any other things i should keep my eye on?

    here is the final result when the first is with the script above and the second is the one i am aiming for

    http://screenshotcomparison.com/comparison/192436

    any suggestions to improve it even more?

    also if i want to stay at 720p and not go up to 1080p how should the script look?
    Last edited by zanzar; 1st Dec 2016 at 11:24.
    Quote Quote  
  17. Santiag() is an antaliaser. It smooths jagged edges. If you don't have jagged edges you don't need to antialias.
    Quote Quote  
  18. Originally Posted by jagabo View Post
    Santiag() is an antaliaser. It smooths jagged edges. If you don't have jagged edges you don't need to antialias.
    ok thanks.

    what about the other part of my last post?
    Quote Quote  
  19. Originally Posted by jagabo View Post
    Santiag() is an antaliaser. It smooths jagged edges. If you don't have jagged edges you don't need to antialias.
    jagabo "McTemporalDenoise" works miracles...
    Last edited by zanzar; 1st Dec 2016 at 16:52.
    Quote Quote  
  20. Originally Posted by jagabo View Post
    Santiag() is an antaliaser. It smooths jagged edges. If you don't have jagged edges you don't need to antialias.
    jagabo any way to deal with the flickering that hysteria sometimes when a there is no black lines dividing but rather different color?

    http://screenshotcomparison.com/comparison/192747

    like in this image,she doesnt move but heir hair changes constantly in a flicking way (like hysteria cant desive weather he should darken it or not causing constant darkening and un-darkening...)


    script:

    Hysteria(strength=6.0, maxchg=26, lowthresh=15, highthresh=20)
    McTemporalDenoise(settings="high").
    nnedi3_rpow2(2, cshift="spline36resize", fwidth=1280, fheight=720).
    Hysteria(strength=5.0, maxchg=28)
    Dehalo_alpha(rx=3, ry=3, darkstr=0.3, brightstr=0.3)
    nnedi3_rpow2(2, cshift="spline36resize", fwidth=1920, fheight=1080)
    Blur(1.0)

    edit: i think the double call for hysteria is doing it ...
    edit:maybe shifting one hysteria() to after the 1080p upscale?
    i guess its hard for him to target stuff like this on lower resolutions

    sample included
    Image Attached Files
    Last edited by zanzar; 4th Dec 2016 at 15:05.
    Quote Quote  
  21. Try increasing lowthresh to about 30. Use it in both calls to Hysteria. Experiment with different values.
    Quote Quote  



Similar Threads