TDeint causing jagged lines in deinterlaced and IVTC'd video

[Vitality]

My code:

MPEG2Source("%source_file%")
TDeint(mode=1, mtnmode=3, blim=100)
LoadPlugin("J:\StaxRip-x64-2.0.0.0\Apps\Plugins\AVS\GRunT\GRunT.dll")
Import("J:\StaxRip-x64-2.0.0.0\Apps\Srestore.avsi")
Srestore(frate=23.976)
KNLMeansCL(D = 1, A = 1, h = 4, device_type="auto")
DeHalo_alpha_mt(rx=2.0, ry=2.0, darkstr=1.0, brightstr=1.0, lowsens=50, highsens=50, ss=1.5)

I use TDeint to deinterlace and SRestore to bring the framerate back to 23.976 instead of 59.940. However there is a weird line effect that isn't preset in the source but in the encode. I would say this is because the source is a mix of hard and soft telecine and this part is only soft-telecined, and TDeint is trying to deinterlace a progressive source.


VOB:

Deinterlaced and IVTC'd video:

[jagabo]

Deinterlacers sometimes misidentify closely spaced horizontal lines as interlacing and kick in a deinterlacer. That's probably what you are seeing. Upload a short sample if you want it examined more closely.

[manono]

I use TDeint to deinterlace and SRestore to bring the framerate back to 23.976

If, as you say, it's a mix of hard and soft telecine, why are you bobbing it followed by using SRestore on it? If you insist on doing it that way, use a better bobber, one such as QTGMC. Or just perform a normal IVTC.

TDeint is trying to deinterlace a progressive source.

It's doing what you told it to do - double-rate deinterlace everything. And while it's not so bad as a bobber, it has a tendency to leave aliasing behind.

Deinterlaced and IVTC'd video:

Where's the IVTC? Not in the script you provided. And if there was an IVTC, you might want to have TFM use a better deinterlacer - QTGMC again - or loosen the CThresh settings, or even do away with the post-processing entirely. That can be risky, though.

As jagabo suggests, a short sample will tell the tale.

[jagabo]

For double frame rate field matching (much faster than QTGMC) try:

Interleave(TFM(field=1), TFM(field=0))

If that gives you the wrong field order swap the field values.

[Vitality]

I use TDeint to deinterlace and SRestore to bring the framerate back to 23.976

If, as you say, it's a mix of hard and soft telecine, why are you bobbing it followed by using SRestore on it? If you insist on doing it that way, use a better bobber, one such as QTGMC. Or just perform a normal IVTC.

TDeint is trying to deinterlace a progressive source.

It's doing what you told it to do - double-rate deinterlace everything. And while it's not so bad as a bobber, it has a tendency to leave aliasing behind.

Deinterlaced and IVTC'd video:

Where's the IVTC? Not in the script you provided. And if there was an IVTC, you might want to have TFM use a better deinterlacer - QTGMC again - or loosen the CThresh settings, or even do away with the post-processing entirely. That can be risky, though.

As jagabo suggests, a short sample will tell the tale.

Doesn't an IVTC filter work by deinterlacing and then removing the duped frames? When I used TFM/TDecimate I got a lot of chroma ghosting. I was told to use SRestore



[Attachment 47826 - Click to enlarge]

[Vitality]

I use TDeint to deinterlace and SRestore to bring the framerate back to 23.976

If, as you say, it's a mix of hard and soft telecine, why are you bobbing it followed by using SRestore on it? If you insist on doing it that way, use a better bobber, one such as QTGMC. Or just perform a normal IVTC.

TDeint is trying to deinterlace a progressive source.

It's doing what you told it to do - double-rate deinterlace everything. And while it's not so bad as a bobber, it has a tendency to leave aliasing behind.

Deinterlaced and IVTC'd video:

Where's the IVTC? Not in the script you provided. And if there was an IVTC, you might want to have TFM use a better deinterlacer - QTGMC again - or loosen the CThresh settings, or even do away with the post-processing entirely. That can be risky, though.

As jagabo suggests, a short sample will tell the tale.

Doesn't an IVTC filter work by deinterlacing and then removing the duped frames? When I used TFM/TDecimate I got a lot of chroma ghosting. I was told to use SRestore. And I was also told to bob it.


VTS_03_1.demuxed.m2v

For double frame rate field matching (much faster than QTGMC) try:

Interleave(TFM(field=1), TFM(field=0))

If that gives you the wrong field order swap the field values.

That gave me some leftover interlacing and the same ghosting issues, but a little bit tamer.

Zoom in, it's very faint:

Isn't this interlaced YV12 being treated as progressive YV12 or vise versa? The source VOB claims to be Progressive scan in MediaInfo, but it has the 2:3 Pulldown flags. It's mix of hard and soft telecine. This scene at the toy factory seems to be hard telecined.

[manono]

Doesn't an IVTC filter work by deinterlacing and then removing the duped frames?

No, an IVTC field matches followed by removing the dupe frames. Field matching (putting the right fields together) is very different from deinterlacing.

If the IVTC doesn't work well - and sometimes it doesn't - then the bob/decimate solution might sometimes work better. But you're using the wrong bobber (TDeint(Mode=1)) if the resulting aliasing bothers you. My suggestion was to try QTGMC as the bobber. It wasn't really designed to work with animations with repeated frames, but you'll never know until you try. jagabo proposed a different solution which, apparently, improves the result. The resulting minor interlacing can be removed by adding Vinverse afterwards or, perhaps, by lowering the CThresh.

[johnmeyer]

Manono, as usual, is correct:

IVTC <> Deinterlacing

[Vitality]

Manono, as usual, is correct:

IVTC <> Deinterlacing

OK so it seems the actual ghosting is part of the actual stream. I watched the VOB with a deinterlace filter (MPC-HC, hardware deinterlace), but when you watch it raw it just looks like interlacing lines.




Is there anywhere to remove the chroma bleed? And if the deinterlace filter is causing the chrome bleed, could I hook up my DVD player over composite/S-Video into my Digital8 camcorder and stream the "raw" NTSC signal to my PC over DV-AVI?

[jagabo]

All of this was covered in your previous thread.
https://forum.videohelp.com/threads/391643-Help-with-telecined-cartoon-that-has-combin...ues-on-encodes

The chroma of your video is screwed up in the source. The luma is screwed up in places too. Problems like this are quite common with cartoons because shots are often crudely time adjusted after being telecined.

[Vitality]

All of this was covered in your previous thread.
https://forum.videohelp.com/threads/391643-Help-with-telecined-cartoon-that-has-combin...ues-on-encodes

The chroma of your video is screwed up in the source. The luma is screwed up in places too. Problems like this are quite common with cartoons because shots are often crudely time adjusted after being telecined.

Is there any sort of post-processing or filter I can do after QTGMC().SRestore()? IIRC I saw a thread here with a different cartoon which had the same sort of issue, and suprisiy that cartoon was the same exact thing, a mix of soft and hard telecine.

[jagabo]

You can always go through and replace bad frames with the frame before, the frame after, or a motion interpolated frame.

[Vitality]

You can always go through and replace bad frames with the frame before, the frame after, or a motion interpolated frame.

So I decided to give Handbrake a shot, and I used Handbrake's Decomb/Bob filter and the Detelecine one. Look at this:




The chroma bleed is almost completely gone. Now, how do i use these filters on AviSynth? HB is open source so it should be easy, right? Or are the Decomb/Detelecine filters already on AviSnyth?

[SaurusX]

Handbrake has some good routines built into it. It's not a slouch. But anything they have is going to be outclassed by what's available for AVISynth. I just think you're not applying the filters in the correct way. TFM will do exactly what Deinterlace-Decomb does(deinterlace only the combed areas), but do it better and in higher quality. Use something like this:

MPEG2SOURCE(...)
i=last
edeint=i.nnedi3(-2)
TFM(pp=7,mode=3,clip2=i.deint(mode=2,edeint=edeint )) #Also try pp=6. Add the "mi" parameter and continue to decrease if you still see residual interlacing
Tdecimate(mode=1)