VideoHelp Forum
+ Reply to Thread
Page 1 of 4
1 2 3 ... LastLast
Results 1 to 30 of 119
Thread
  1. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Hi folks. So I've made tremendous progress on my toons. I've been working with TemporalDegrain at reasonable settings, throwing in a dash of TTempSmooth (just a dash) and then shipping it off to NeatVideo in V-dub for the finishing touch. So far, I've had great results (sanlyn, small settings with multiple filters---great advice ). My details are kept intact, no ghosts, and my cleaning is going superb. Except for one teensy thing (always something, isn't it?). Grid artifacts.

    Now before we go nuts, these aren't too visible. You have to really look at the screen, and they are only noticeable in dark scenes (hence my attached clip). I know that FFT3D produces these artifacts, and, if the filter is used alone, there are settings to lower it. Except that in this case, I'm working with TemporalDegrain, which uses the filter. Problem is, I'm not sure if TemporalDegrain gives me the option to tamper with FFT3D settings within the TemporalDegrain script, as much as i would be allowed to do so with the filter in and of itself.

    I'm also having the suspicion that TTEmpSmooth may have something to do with this as well, as I've seen these artifacts on previous projects where I've used TTempSmooth and not TemporalDegrain nor FFT3d.

    So, attached is the clip. Look at the background closely and you'll see a faintly thin grid all over the background. Any way to get rid of this WITHOUT disrupting the rest of the functions? My script is perfect and I wouldn't want to undo anything else.

    Here's my script:

    Code:
    Import("C:\Program Files\AviSynth 2.5\plugins\TemporalDegrain.avs")
    MPEG2Source("C:\Users\joiknoff\Desktop\New Masters\d2vfiles\doodoohead.d2v", cpu=3)
    TFM(d2v="C:\Users\joiknoff\Desktop\New Masters\d2vfiles\doodoohead.d2v")
    TDecimate(mode=1)
    Crop(6, 0, -6, -0)
    Tweak(hue=0.0, sat=0.85)
    TemporalDegrain(sad1=200, sad2=100, pel=2)
    TTempSmooth(maxr=5, lthresh=4, cthresh=3,interlaced=false)
    AddBorders(6, 0, 6, 0)
    Thanks.
    Attached Files
    Last edited by unclescoob; 11th Mar 2012 at 21:27.
    Quote Quote  
  2. Member
    Join Date: Sep 2007
    Location: Canada
    Search Comp PM
    If you encounter gridding artifacts, then try one or more of the following:
    • Increase ow and/or oh
    • Increase bw and/or bh
    • Decrease sigma



    http://www.aquilinestudios.org/avsfilters/spatiotemp.html#fft



    Problem is, I'm not sure if TemporalDegrain gives me the option to tamper with FFT3D settings within the TemporalDegrain script, as much as i would be allowed to do so with the filter in and of itself.
    Yes it does, If you look at TemporalDegrain.avs, the settings (and their default values) are listed
    Quote Quote  
  3. Banned
    Join Date: Oct 2004
    Location: New York, US
    Search Comp PM
    No "grid" pattern in the sample posted. Plenty of macroblocks and high-compression artifacts. Note that the "pattern" you mention isn't a grid, it's blocky, chunky, low-frequency clumping. It occurs mostly in dark objects, seldom seen in brighter parts of the image -- inidcates a birtrate or matrix used that allocated very few bitrates to dark areas. What kind of GOP structure is this m2v? The GOP setup changes every few frames, isn't consistent at all. Can't say I've seen that before (but, then, there's a lot I haven't seen).

    Temporal smoothers won't do much here. Try DeBlock and GradFunDBMod. Try alternating the filter sequence, use one or both of these before calling temporaldegrain. It might help deblockers and smoothers a bit if your keep your darks above RGB16 (can always darken later). Clipped colors are difficult to filter.

    Have no sample of the unprocessed original source, but if you look you'll see the same noise is there before applying any filters.
    Last edited by sanlyn; 23rd Mar 2014 at 11:38.
    Quote Quote  
  4. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Actually that clip was waaaaaaay noisier than this. It looks great on my tv. We need to keep in mind that the computer monitor always shows more junk than the actual TV.

    I encode with HcEnc. The GOP is 12 (pulldown applied at encoding). Should I raise it to 14?

    Lumgain 3 (for dark areas) and AQ is set to high (precisely for allocating more bits to flat areas).

    So what you're saying is, before I apply any filtering, I should change the colorspace? What colorspace should I set it to?
    Quote Quote  
  5. Member themaster1's Avatar
    Join Date: Nov 2006
    Location: France
    Search Comp PM
    I made a topic about hc-enc, you should read here imo
    lum gain:1 , Aq:2, matrix : avamat6 seems best in many situations
    12 gop for pulldown seems right
    Quote Quote  
  6. Banned
    Join Date: Oct 2004
    Location: New York, US
    Search Comp PM
    Originally Posted by unclescoob View Post
    Actually that clip was waaaaaaay noisier than this. It looks great on my tv. We need to keep in mind that the computer monitor always shows more junk than the actual TV.

    I encode with HcEnc. The GOP is 12 (pulldown applied at encoding). Should I raise it to 14?

    Lumgain 3 (for dark areas) and AQ is set to high (precisely for allocating more bits to flat areas).

    So what you're saying is, before I apply any filtering, I should change the colorspace? What colorspace should I set it to?
    The plugins you're using want YV12, no need to change it. GOP and colorspace are two different issues. You're using degrainers and smoothers to tackle grain, which works, but it doesn't do much for the kind of compression artifacts in the original (which we haven't seen) or the sample.

    I don't use HCenc, nor do I frameserve between apps. It takes too long to tweak processing that way. Since I don't know HCenc, maybe its behavior is normal. Anyway, I've no problem playing the sample. But if you're asking for a 12-frame GOP, you don't appear to be getting it. At least, my MPEG Editor says you're getting something else:

    GOP seen in TMPGenc Editor v3, GOP set for 12-frame set. Numbers show frames in each GOP:
    Attachment 11434

    GOP seen in TMPGenc Editor v3, GOP set for 16-frame set (3:2 pulldown). Encoded with TMPGenc Plus. Numbers show frames in each GOP:
    Attachment 11435

    Gspot image of presentation order of frames in Sample1.m2v shows irregular GOP structure of I,P,B frames:
    Attachment 11436

    Gspot image of presentation order of frames in 16-frame GOP, show consistent structure of I,P,B frames:
    Attachment 11437

    I have no idea if this makes any difference. In any case, this has little to do with cleaning up artifacts. The HCenc matrix you quote might be allocated plenty of bits to dark areas for a new encode, but I'm talking about the original source.
    Last edited by sanlyn; 23rd Mar 2014 at 11:38.
    Quote Quote  
  7. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Originally Posted by themaster1 View Post
    I made a topic about hc-enc, you should read here imo
    Where's the link?

    Sanlyn, I'll post the original source tonight. Deblock and Gradfun....will they kill details in the process of doing what they're supposed to do?
    Quote Quote  
  8. Banned
    Join Date: Oct 2004
    Location: New York, US
    Search Comp PM
    You have to magnify the image somewhat to see any softening, but if it matters you can sharpen anime nicely with LSFMod or NeatVideo. LSFMod uses MaskTools and RemoveGrain, which you already have. It's an avsi script: copy the whole avsi file into Avisynth plugins to load it automatically. Don't fiddle with LSFMod's "smode" settings; if you do, you'll spend all day loading two more biggies (warpsharp and VariableBlur). Just use LSFMod's "Strength" setting, the defaults don't require the last two plugins. The Strength default = 100. I'd start with 50 ("LSFMod(Strength=50)". But if you run NeatVideo, its sharpener will do the trick -- if you need it. A sample of LSFMod in action is on this page (scroll down to "sharpeners"): http://www.animemusicvideos.org/guides/avtech31/post-qual.html

    The sample posted has mosquito noise around edges. Use sharpeners last, or you'll end up sharpening that noise!
    Last edited by sanlyn; 23rd Mar 2014 at 11:39.
    Quote Quote  
  9. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Originally Posted by sanlyn View Post
    Use sharpeners last, or you'll end up sharpening that noise!
    Fine. Except TemporalDegrain has a sharpener IN the script. But from what I read, it's a slight sharpener. Any ideas?

    So what will gradbfun do for the artifacts?
    Quote Quote  
  10. Member themaster1's Avatar
    Join Date: Nov 2006
    Location: France
    Search Comp PM
    Originally Posted by unclescoob View Post
    Originally Posted by themaster1 View Post
    I made a topic about hc-enc, you should read here imo
    Where's the link?
    see Here
    Quote Quote  
  11. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    I don't know much about GOP but I don't like the way your editor is showing my frames. They're all discombobulated and out of place! Geez man, it seems like just when I find the solution to something, another problem pops up.
    Quote Quote  
  12. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    So you're saying that Avamat6 keeps more grain and reduces artifacts in dark scenes. Does it ruin the bright scenes? Do you think encoding with avamat will help? I'll try it at home. Also, Master....how do I fix this GOP structure that sanlyn's program is showing???

    I IVTC my NTSC episode with avisynth and then applied pulldown at a 12 GOP with HcEnc when encoding. Where did I go wrong?
    Quote Quote  
  13. Banned
    Join Date: Oct 2004
    Location: New York, US
    Search Comp PM
    I don't know that the GOP thing makes a difference. It played OK for me, looks alright in Avisynth AVI. The first few frames get jumbled (that happened with your previous 'toon as well). I just converted m2v to AVI, then reencoded (looks worse, of course, with another encode) but looked OK. I just don't know what HCenc is doing there, or if it even matters. Just looked weird.

    Re-encoded, GOP of 12 frames each group:
    Attachment 11441

    GradFun2dbMod is used for block noise and banding issues. DeBlock works too.

    TemporalDegrain does have a sharpener, but it's not that sharp. Basically it's optimized to put out what you put in as far as sharpness goes. Thus, it's a degrainer, not a smoother.
    Last edited by sanlyn; 23rd Mar 2014 at 11:39.
    Quote Quote  
  14. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Yeah but if my GOP is looking like it did before, that may not be a good thing. so all you did was re-encode to MPEG2 and it now looks good. What program did you use to encode it? What settings did you use? You said you don't use HcEnc. What encoder do you use, and at what settings? (I know, same question written twice, sorry)
    Quote Quote  
  15. Member
    Join Date: Sep 2007
    Location: Canada
    Search Comp PM
    1) You can see the grid pattern if you raise the gamma. Personally, I wouln't worry to much about it. On a calibrated setup, you shouldn't be seeing them. But that doesn't mean other viewers have calibrated setups

    You can either raise gamma (e.g. using levels, or adjust your monitor or graphics card) or use histogram("luma") to visualize . People use that to optimize their encodes so they know what to adjust . It's probably overkill, but if it looks good with histogram("luma") then the Y' signal should look good under any conditions or setup

    2) Fixed GOP length or structure is not necessarily a good thing. It should be content dependent. There might more efficient ways of encoding

    eg. - a quick scene change or flashes, explosions , where "B" frame was supposed to be placed . It might be better to insert a high bitrate "I" frame there instead in that situation

    But a simple scene where there are not changes (e.g. maybe a slow pan), then a set fixed GOP might be ok in that situation
    Attached Images  
    Quote Quote  
  16. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    PDR - what settings do you suggest for my GOP structure then?
    Quote Quote  
  17. Member
    Join Date: Sep 2007
    Location: Canada
    Search Comp PM
    Originally Posted by unclescoob View Post
    PDR - what settings do you suggest for my GOP structure then?
    I would leave it as is, unless you see a specific reason to change it. Eitherway, the GOP settings are not the reason for the grid artifacts
    Quote Quote  
  18. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    I was advised also to change the block size in the FFT3D filter on TemporalDegrain, and/or the Sigma levels. My goal is a professional output and I can't rely on monitor calibration. I don't want the grids in the picture, period. But thanks for showing me that though.
    Quote Quote  
  19. Banned
    Join Date: Oct 2004
    Location: New York, US
    Search Comp PM
    Originally Posted by unclescoob View Post
    Yeah but if my GOP is looking like it did before, that may not be a good thing. so all you did was re-encode to MPEG2 and it now looks good. What program did you use to encode it? What settings did you use? You said you don't use HcEnc. What encoder do you use, and at what settings? (I know, same question written twice, sorry)
    TMPGenc Plus 2.5. I use settings that are common to most MPEG1/2 encoders. Comes with 2 matrices (default and MPEG standard). You can add others, but I never felt the need.

    Thanks for the GOP info, poisondeathray. The weird appearance in editors didn't seem to affect playback.

    The "grid" pattern (like the fabric on my sofa!) looks like compression artifacts to me IMHO, from the source. I always have a devil of a time getting rid of it.

    Originally Posted by unclescoob View Post
    My goal is a professional output and I can't rely on monitor calibration.
    Why not?
    Last edited by sanlyn; 23rd Mar 2014 at 11:40.
    Quote Quote  
  20. Member
    Join Date: Sep 2007
    Location: Canada
    Search Comp PM
    Originally Posted by sanlyn View Post

    Thanks for the GOP info, poisondeathray. The weird appearance in editors didn't seem to affect playback.
    In most editors, the GOP structure doesn't matter, because they are decoded and decompressed - so everything is treated as uncompressed frames

    But when you use other compression types than DVD compliant MPEG2, (Where there might be enormous GOP sizes), that it might affect editor performance (longer seek latency)

    The "grid" pattern (like the fabric on my sofa!) looks like compression artifacts to me IMHO, from the source. I always have a devil of a time getting rid of it.
    It might be, but we don't have the source so without more information we are just guessing....



    PS Isn't this available on retail DVD?
    Quote Quote  
  21. Member 2Bdecided's Avatar
    Join Date: Nov 2007
    Location: United Kingdom
    Search Comp PM
    It's far better to have an encoder that adapts the GOP structure to the content, than one that sticks with a dumb fixed pattern.

    e.g. It makes good sense to use an I-frame on any scene change. An encoder that uses a fixed-GOP pattern may use a B-frame on a scene change. Problem is, a B-frame only describes the differences between a previous frame and the current one, typically for the sake of using fewer bits because differences are often small. How dumb would it be to use this type of encoding on a frame that's completely different from all the previous ones?! Yet that's exactly what fixed-GOP encoding will do.

    A good encoder will vary the GOP structure while remaining DVD compliant, and every decoder+player in the world will play such a stream just fine.

    Cheers,
    David.
    Quote Quote  
  22. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Like HcEnc, right?
    Quote Quote  
  23. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    How do you guys feel TMPGenc compares to HcEnc, by the way? (not changing the subject here, just curious)
    Quote Quote  
  24. Member themaster1's Avatar
    Join Date: Nov 2006
    Location: France
    Search Comp PM
    As far as i know tmpgenc still convert to rgb whatever video you throw to it (= color accuracy pb), with hcenc you have control over everything so you do the maths
    Quote Quote  
  25. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Originally Posted by sanlyn View Post
    The "grid" pattern (like the fabric on my sofa!) looks like compression artifacts to me IMHO, from the source. I always have a devil of a time getting rid of it.
    My grid patterns are due to the filter. They weren't there when I used other filters. It's definitely FFT3d, but I will increase the block size and see what happens.

    Originally Posted by unclescoob View Post
    My goal is a professional output and I can't rely on monitor calibration.
    Originally Posted by sanlyn View Post
    Why not? Seems a rather odd thing to say.
    What I mean is, I do not want the artifacts in my encode at all. I would not want to rely on someone else's monitor calibration in hopes that the screen doesn't display them. I just don't want them in my clip.
    Quote Quote  
  26. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Originally Posted by poisondeathray View Post
    PS Isn't this available on retail DVD?
    Yeah. The episodes that I'm working on ARE the original DVD rips. And we can see what "fine" job the production company did with the DVD set

    I'll show you a source of the original clip tonight when I get home. These people just dumped the taped episodes on DVDs, wrapped it up, put a price tag on it and called it a day.

    The best thing about it is the fact that I have gotten to learn about video a bit. They should have done a better job at cleaning these episodes.
    Quote Quote  
  27. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Ok, well I have to re-rip that episode into my computer (only have the d2v file, which I don't think I can load here). But here's a clip of a dark scene from a different episode.
    Attached Files
    Quote Quote  
  28. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    What does this error message require me to do with Deblock?

    "width and height must be mod 8"
    Quote Quote  
  29. Originally Posted by unclescoob View Post
    What does this error message require me to do with Deblock?

    "width and height must be mod 8"
    Both the height and width (of the video being fed into Deblock) must be divisible by 8.

    ...only have the d2v file, which I don't think I can load here
    And which would be useless anyway to someone looking for a sample.
    Quote Quote  
  30. Member
    Join Date: Dec 2010
    Location: New York
    Search Comp PM
    Ah. Gotcha. Thanks.
    Quote Quote  



Similar Threads