Help with proper decombing or deinterlacing

[killerteengohan]

I have an anime source I have been trying to get to come out right. I think it's supposed to be 29.970fps. The Japanese DVD is 29.970fps Progressive, and the Japanese Blu-Ray is also 29.970fps progressive. The DVD's I have in my country appear to be combed or interlaced.

This appears to have quite a bit of blending in it too.

I have been using the usual tfm(order=-1, slow=2) that I use for most anime's, but for some reason it comes out kind of jumpy, instead of smooth panning in some places. It looks like it's bouncing along back and forth instead of a single direction glide at certain parts. Most of it appears perfectly fine, but some parts are ruined looking. Playing back the VOB before encoding doesn't have this issue.

Look at the 5:32 - 5:35 area. It seems to affect that part a lot more than most others. That part comes out terrible for me.

Does this require some kind of special treatment or pulldown flags? So far I cant figure this source out. I think it may possibly be a hybrid where some is best at 29.970fps and other parts are better at 23.976fps.


Would someone have a look and share some suggestions please?

Here is the file I am working with.

https://mega.nz/file/tg50lQrD#yjMk0NANANnj44CCz0Ol28Re7R9YnlQN1Y19o6GTSCg

[killerteengohan]

This is the results I keep getting with the tfm.

https://mega.nz/file/IgZHkaDT#EEdQkIkTWbYNs7dCyetZDvzRxILJaXAMOWreUfpbb28

Look at the 5:32 - 5:35 area. It has a really weird wobble that's not nearly as obvious when the VOB, or disc is played.

[jagabo]

QTGMC().SRestore(frate=23.976)

Though other parts of the video appear to be 29.97p.

[killerteengohan]

QTGMC().SRestore(frate=23.976)

Though other parts of the video appear to be 29.97p.

Yeah, I think the CGI is 29.97 and most of the other animation is 23.976. I didn't like how 23.976 kind of ruined those parts that appear to play best at 29.97 with jittery motion.
I was hoping I could get 29.97 for the entire thing. I don't have the time and patience for a series this long to make vfr timecodes files for each one. Unless perhaps you have an easy way of finding only the 29.97 parts to get the frame numbers a lot quicker. Only way I would be able to find them is by watching the video from beginning to end and making notes of the times it jitters, which would take a long time.

What makes you suggest 23.976 over 29.970 for this source material?

[Selur]

You might want to try ExBlend (https://forum.doom9.org/showthread.php?t=175948) for the blend removal,...

[jagabo]

If you want 29.97 fps you can SRestore(frate=29.97). Of course, the 23.976 fps segments will be jerky.

[killerteengohan]

If you want 29.97 fps you can SRestore(frate=29.97). Of course, the 23.976 fps segments will be jerky.

Yeah, that's one problem I had with it. I tried it before I asked the original question.

QTGMC can output 60fps after deinterlacing. I am curious, can TFM do that as well? I don't think so, but I thought I would ask.

[jagabo]

Yes, TFM can be used to bob 30i fps to 60p. Try one of these:

# smart bob a video using TFM

function TFMBob(clip v, int "pp", int "cthresh")
{
    pp = default(pp, 6)
    cthresh=default(cthresh, 9)

    v.GetParity() ? Interleave(TFM(v, field=1, pp=pp, cthresh=cthresh), TFM(v, field=0, pp=pp, cthresh=cthresh)) : Interleave(TFM(v, field=0, pp=pp), TFM(v, field=1, pp=pp, cthresh=cthresh))
}


# using QTGMC for clip2

function TFMBobQ(clip v, int "pp", int "cthresh")
{
    pp = default(pp, 6)
    cthresh=default(cthresh, 9)

    q = QTGMC(v)
    qe = q.SelectEven()
    qo = q.SelectOdd()
    v.GetParity() ? Interleave(TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh), TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh)) : Interleave(TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh), TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh))

}


# using nnedi3 for clip2

function TFMBobN(clip v, int "pp", int "cthresh")
{
    pp = default(pp, 6)
    cthresh=default(cthresh, 9)

    n = nnedi3(v, field=-2, nns=4)
    ne = n.SelectEven()
    no = n.SelectOdd()

    v.GetParity() ? Interleave(TFM(v, field=1, clip2=ne, pp=pp, cthresh=cthresh), TFM(v, field=0, clip2=no, pp=pp, cthresh=cthresh)) : Interleave(TFM(v, field=0, clip2=no, pp=pp, cthresh=cthresh), TFM(v, field=1, clip2=ne, pp=pp, cthresh=cthresh))
}

Use it something like:

Mpeg2Source("VTS_07_1.d2v", CPU2="ooooxx", Info=3) 
TFMBobN()
SRestore(frate=29.97)
Stab(mirror=15)
Crop(8,0,-8,-0)

[killerteengohan]

Thanks! I will give that a try tomorrow when I get home and let you know what I think.

I was asking because for some reason, every time I use QTGMC instead of TFM for any source, it makes lines, and details more blurry, or out of place. Have a look at this comparison. It's the same frame. Because of this, I try to use QTGMC as a last resort only.

https://slow.pics/c/bJHRdg8S

I hope the TFMBob does a better job.

[jagabo]

That's pretty much to be expected as QTGMC attempts to smooth away residual aliasing and buzzing edges.

[killerteengohan]

That's pretty much to be expected as QTGMC attempts to smooth away residual aliasing and buzzing edges.

Can that be disabled in QTGMC? Aside from that one issue that makes me hate it, I like it overall.

Thanks for showing me that TFMBob exists. I tried the TFMBobN. It looks like it does do MUCH better than QTGMC with the lines and details. I may never have to use QTGMC again lol, unless the aliasing smoothing really helps.

Does the TFMBob have a function similar to QTGMC's fpsdivisor=2? I think the fpsdivisor=2 is throwing out every second frame. Any other alternatives I can try besides srestore to bring it back to 29.970fps? I know there's changefps, but I hate that one. If tdecimate can do it, I'm not 100% sure how to set that one up for 60 to 29.97 or 60 to 23.976. I have only used tdecimate to bring 29.970 down to 23.976.

[johnmeyer]

Can that be disabled in QTGMC?

From the QTGMC documentation:

Noise Bypass / Denoising

[jagabo]

Keep in mind that TFM is a field matcher, not a deinterlacer. When there's no matching field it falls back on rather simple deinterlacing. Providing a clip2 helps a bit. So TFM is not appropriate for true interlaced material (camcorder video, live sports, TV news, etc.). My TFMBob functions sometimes work for PAL/NTSC conversions.

The TFM equivalent of QTGMC's fpsdivisor=2 is simply to call TFM only once with TFM(field=0) or TFM(field=1). The TFMBob filters call TFM twice, once with field=0 and once with field=1, then interleaves the two results.

For your purposes here you want to use SRestore. It preferentially discards the frames with the most blending. TDecimate preferentially discards duplicate frames (blended frames are not seen as duplicates). And ChangeFPS blindly discards every Nth frame (every other frame when going from 60p to 30p).

[killerteengohan]

I always use clip2 with TFM and QTGMC ever since you taught me about it a long time back, because it does help and I have used it a lot to catch more leftovers TFM misses. I tried using TFMBobQ so I could use QTGMC as clip2, and its ruining the video like this. Am I using it wrong?
https://forum.videohelp.com/images/imgfiles/noLkpLD.png
https://forum.videohelp.com/images/imgfiles/sJL8GsE.png

TFMBobN doesn't seem to have a clip2 parameter.

[jagabo]

I always use clip2 with TFM and QTGMC ever since you taught me about it a long time back, because it does help and I have used it a lot to catch more leftovers TFM misses. I tried using TFMBobQ so I could use QTGMC as clip2, and its ruining the video like this. Am I using it wrong?
https://forum.videohelp.com/images/imgfiles/noLkpLD.png
https://forum.videohelp.com/images/imgfiles/sJL8GsE.png

I haven't seen it do that but it looks like some kind of chroma subsampling error. Make sure the field order is set correctly before calling it. Also make sure it's one of the supported color formats. Try ConvertToYV12(interlaced=true).

TFMBobN doesn't seem to have a clip2 parameter.

Both TFMBobQ and TFMBobN create clip2 from the input clip. The former using QTGMC, the latter using nnedi3.

[rrats]

I think there had been a thread discussing blend fix the Initial D anime series. It suffers from the same blending as many other 90s anime (like the bloody evnagleion). As far as I know, the artefact is not fully fixable. It was probably a low quality anti flicker filter reducing the flicker caused by differences in light (50Hz) and video system (60Hz) in a telecine machine.


[Attachment 67116 - Click to enlarge]

AnimeIVTC might fix some motionless stills or the lucky scenes animated on thirds, but the blending will be present.

Does this require some kind of special treatment or pulldown flags? So far I cant figure this source out. I think it may possibly be a hybrid where some is best at 29.970fps and other parts are better at 23.976fps.

pretty sure animeivtc's mode 4 (hybrid) can help

[killerteengohan]

I haven't seen it do that but it looks like some kind of chroma subsampling error. Make sure the field order is set correctly before calling it. Also make sure it's one of the supported color formats. Try ConvertToYV12(interlaced=true).

I'm not sure whats with the purple like in those images. It comes out fine if I encode it, and it doesn't do that in my other avisynth previewing software. None of those suggestions were a fix.

I guess the program I have been using all this time has suddenly decided it doesn't like QTGMC anymore lol. I restarted the program, and it went away. After using it for a while, that started happening again. Apparently its nothing to worry about and is a simple software issue.

Thanks for the TFMBob information. If I come across any problems or questions, I will ask.

[jagabo]

Many programs have problems with MPEG 2 in MKV. That's why I remuxed to VOB (MPEG 2 program stream) first, then used DgIndex/Mpeg2Source() to read the video.

[killerteengohan]

Many programs have problems with MPEG 2 in MKV. That's why I remuxed to VOB (MPEG 2 program stream) first, then used DgIndex/Mpeg2Source() to read the video.

I'm using the original VOB I extracted from the disc that's been file indexed with DGIndex. That was pretty much a one time issue for me. I think it's because I had multiple preview tabs with different scripts all trying to access the same source file at the same time. Closing and re-opening fixed it until using a bunch of preview tabs again. It's funny that only the QTGMC filter set that off when it was happening.

[jagabo]

It sounds like something isn't completely thread safe yet.

[killerteengohan]

I have a question about the TFMBobQ filter.

# using QTGMC for clip2

function TFMBobQ(clip v, int "pp", int "cthresh")
{
    pp = default(pp, 6)
    cthresh=default(cthresh, 9)

    q = QTGMC(v)
    qe = q.SelectEven()
    qo = q.SelectOdd()
    v.GetParity() ? Interleave(TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh), TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh)) : Interleave(TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh), TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh))

}

Is that using QTGMC at it's default settings? Can I alter the QTGMC settings in the code?

I would like it to be QTGMC with parameters like these.
QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", FPSDivisor=2, sourcematch=3, sharpness=0.1)

[jagabo]

I have a question about the TFMBobQ filter.

# using QTGMC for clip2

function TFMBobQ(clip v, int "pp", int "cthresh")
{
    pp = default(pp, 6)
    cthresh=default(cthresh, 9)

    q = QTGMC(v)
    qe = q.SelectEven()
    qo = q.SelectOdd()
    v.GetParity() ? Interleave(TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh), TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh)) : Interleave(TFM(v, field=0, clip2=qo, pp=pp, cthresh=cthresh), TFM(v, field=1, clip2=qe, pp=pp, cthresh=cthresh))

}

Is that using QTGMC at it's default settings?

Yes.

Can I alter the QTGMC settings in the code?

Yes.

I would like it to be QTGMC with parameters like these.
QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", FPSDivisor=2, sourcematch=3, sharpness=0.1)

You can't use FPSDivisor=2. The number of frames from QTGMC and TFMBobQ must match.

[killerteengohan]

How do you alter the avsi so it's using this?
QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, sharpness=0.1)

[jagabo]

Just add all the parameters to the QTGMC call.

[killerteengohan]

Thanks

[killerteengohan]

TFM with QTGMC for clip2 is coming out different than TFMBobQ with QTGMC as clip2.

Have a look.
https://slow.pics/c/bMIkTbug

The regular TFM with clip2 is catching those leftover combing/interlacing artifacs so much better for some reason.
If TFMBobQ is essentially also QTGMC for clip2, why is it doing a worse job at this part?


Here is how I am using it

tfm(order=-1, slow=2, cthresh=8, clip2=QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", fpsdivisor=2, sourcematch=3, sharpness=0.1))
srestore(frate=23.976)

and

TFMBobQ(cthresh=8)
SRestore(frate=23.976)

If it helps any, heres the original VOB I am working with.
https://mega.nz/file/VwB1WJ5D#yjMk0NANANnj44CCz0Ol28Re7R9YnlQN1Y19o6GTSCg

The part giving me the noticeable issues is about 22:01 - 22:07

[killerteengohan]

So far the only thing I could find to stop that in TFMBobQ is set the cthresh way down to 2. That's making everything else appear softer and is adding in line blends from other frames to other motion parts unfortunately.

[lordsmurf]

I was hoping I could get 29.97 for the entire thing.

You have to edit apart the clips, then process each properly.

I don't have the time and patience

Video requires time and patience (and often budget). There's a reason projects can takes months or years, even decades in a few cases. Video is not a task for the impatient or lazy.

[jagabo]

If you look at the output of TFM(clip2=QTGMC()) vs. TFMBobQ().SelectEven() you'll see they are nearly identical (as they should be). With this particular video the output of TFMBobQ().SelectOdd() produces many frames that don't match well (aliasing). With all the extra frames (and bouncing) produced by TFMBobQ() SRestore() is having a hard time differentiating blended and original frames. It ends up picking many of the aliased frames. Specifying mode=0 to the TFM() calls in TFMBobQ() might help.

[killerteengohan]

If you look at the output of TFM(clip2=QTGMC()) vs. TFMBobQ().SelectEven() you'll see they are nearly identical (as they should be). With this particular video the output of TFMBobQ().SelectOdd() produces many frames that don't match well (aliasing). With all the extra frames (and bouncing) produced by TFMBobQ() SRestore() is having a hard time differentiating blended and original frames. It ends up picking many of the aliased frames. Specifying mode=0 to the TFM() calls in TFMBobQ() might help.

That makes quite a bit more sense after reading your description.

Actually, the mode=0 seems to have done the trick as well. I'm not 100% sure why it worked, but it looks like it has. I have never used mode=0 before, I have always used the default of mode=1.

0 = 2-way match (p/c)
1 = 2-way match + 3rd match on combed (p/c + n)

Whats with the n, and is mode=1 supposed to be better somehow?