Question about the Right Use of tdecimate

[Womb4T]

Hello,

I'm trying to back-up a NTSC Dvd (using xvid megui encoder and an avs) and I have met a patern that I have never seen before.

The dvd isn't interlaced but every 6 frames, the 6 th is a duplicate of 5 th.

I suppose I should decimate like that :

LoadPlugin("C:\MeGUI\tools\dgindex\DGDecode.dll")
DGDecode_mpeg2source("C:\Movie\test.d2v", info=3)
LoadPlugin("C:\MeGUI\tools\avisynth_plugin\TIVTC.d ll")
TDecimate(cycleR=1,cycle=6)

Do you think I am right ? The final framerate 24.975 surprise me, I have never seen a thing like that.

Thanks for reading me

[PuzZLeR]

Your DvD sounds like a bad transfer of PAL Source -> NTSC DvD. I don't have a sample of this, could you provide us with a small 10s clip?

In the meantime, if you have no multiple frames and smooth movement in your result, you can try and add at the end of your script:

AssumeFPS(23.976)

This will not change the frame count, only the speed. You will also have to adjust the audio by that proportion too.

[Womb4T]

Yes I can , what do you need ? vob or m2v ? Where do I upload it ? Any prefered site ?

[jagabo]

Yes, one duplicate every six frames is from a bad PAL to NTSC conversion. As you surmised, TDecimate(cycleR=1,cycle=6) will discard one of every six frames and leave you 24.975 fps. You might also try TDecimate(mode=2, rate=25) to get exactly 25 fps.

[PuzZLeR]

Ahh, since I'm from NTSC-land I was assuming creating a film 24p result. The O/P now has choices. However, being that it's Xvid, the choice may not matter as much (PAL or NTSC) as it would if he/she were encoding to MPEG-2/DvD.

Womb4T: You can upload a small sample with the "Upload files/Manage attachments" button below displayed as you're posting. However, I don't think you have to. Hopefully it seems pretty clear to you now.

[manono]

Ahh, since I'm from NTSC-land I was assuming creating a film 24p result.

I'd add an AssumeFPS(23.976) at the end of the script also, bring it back to film speed, and then slow the audio to get it back in pitch.

[Womb4T]

@ PuzZLer, jagabo and manono : Thanks for you answers , you trully helped me. I try to improve my skills in encoding and with this one I had reach my limits ^^

As requested, here comes a sample.

I still have two questions :

1) Everyody agrees that it's a bad PAL > NTSC conversion. (I am not surprised, that sort of "movie" is often cheaply authored )

what would be a "good conversion" ? what would be the patern ?

2) Considering the use of AssumeFPS(23.976) I am wondering about streching the audio : Isn't there a risk to denaturate it ?

again, a big thanks for the time you spent answering me.

[jagabo]

1) Everyody agrees that it's a bad PAL > NTSC conversion. (I am not surprised, that sort of "movie" is often cheaply authored )

what would be a "good conversion" ? what would be the patern ?

Encode at 25 fps, apply 3:2:3:2:2 pulldown flags. Or slow it down to 23.976 fps and apply 3:2 pulldown flags.

2) Considering the use of AssumeFPS(23.976) I am wondering about streching the audio : Isn't there a risk to denaturate it ?

Yes, and it's more time/work. That's why I encode at 25 fps and use pulldown flags.

I just got your sample. It has a lot more wrong with it than just duplicate frames. There are tons of blending artifacts too. Is the whole movie like that?

[manono]

1) Usually they're slowed down from 25fps back to film speed. No duplicate frames. Or, assuming the source originated in film to begin with, prepare a proper NTSC master rather than using an already created PAL one.
2) I have no idea what 'denaturate' means. One reason to do it, though, is to get the audio back in tune. In the process of speeding 24fps film to 25fps PAL, the audio winds up playing faster and being a semi-tone higher. If you use a PAL master for the NTSC DVD, it remains speeded up.

Just got the sample. Unfortunately, removing the duplicate frames only goes part of the way towards fixing it. One would need a better sample, one showing more movement, but I don't think it can be entirely fixed at this point. It looks double-blended and deinterlaced to me.

[Womb4T]

Yep the movie is like that, but I think it's intentionnal : obviously they searched to create " vintage effect"...

Another movie, same problem. Another sample.

@ manono : sorry frenglish I wanted to say "will this tech change the sound ?" and got acknowledgement by your answer.

[PuzZLeR]

Bonjour, got your sample too, merci - back to work for now but I will look at it later today. (But I think speaking in French, although I could, may be against the Forum rules.)

Nevertheless I seriously doubt I can add anything further to what Jagabo and Manono already posted on.

[jagabo]

That last clip just has a duplicate every 6 frames but no ghosting. TDecimate should clean it up.

[PuzZLeR]

Darn - just when I was cooking up a script for the first clip...

Yes, this clip is straightforward, and our earlier suggestions should suffice.

But just for the record, about that first clip, I started experimenting with Bob(), Unblend(), and, of course, TDecimate(). Kind of tricky.

[Womb4T]

PuZzler don't be fooled by the first sample : authoring the dvd, they wanted to get a "vintage movie look" (like a silent one for example). What you could take for encoding artifacts maybe intentionnal.

I tested TDecimate(cycleR=1,cycle=6) on the second one, and, honestly, I didn't see the difference with TDecimate(cycleR=1,cycle=5) that megui recommanded ! My eyes are too bad ! (or not enough trained...)

So, I asked a friend to watch and he says that he notices the difference, that TDecimate(cycleR=1,cycle=6) runs smoother. I didn't think that a human eyes can see a difference of one frame by second !

Otherwise, the difference is clearly noticeable during the end credits : TDecimate(cycleR=1,cycle=5) provokes "small jumps" when TDecimate(cycleR=1,cycle=6) perfectly matches the original dvd and runs smooth.

Still got a question : Reading that on http://avisynth.org.ru/docs/english/externalfilters/tivtc_tdecimate.htm)

"Mode 0 = Straight M-in-N decimation. TDecimate will examine each set of N frames and decimate the M most similar frames. The values of M and N are controlled via the cycleR and cycle parameters. Blend decimation of video (30p) is supported in the this mode (hybrid = 1). "

How can I know if the script takes "the right set of 6" (ie the set including the original frame and the duplicate) ?

[jagabo]

PuZzler don't be fooled by the first sample : authoring the dvd, they wanted to get a "vintage movie look" (like a silent one for example). What you could take for encoding artifacts maybe intentionnal.

I tested TDecimate(cycleR=1,cycle=6) on the second one, and, honestly, I didn't see the difference with TDecimate(cycleR=1,cycle=5) that megui recommanded ! My eyes are too bad ! (or not enough trained...)

The motions in your sample are so slow and small it's hard to see. But in the first few seconds, as the car moves to the right, notice the small jumps a few times a second, when cycle=5 is used.

[Womb4T]

Yes, You are right.

Got another problem with another dvd (encoding bad PAL > NTSC conversion is hard !) should I start a new thread ?

Summing up, I have juste met a" new structure"

4 interlaced frames followed by 2 progressives. I have often met 5/1. I used srestore and got nice result... But this time It doesnn't work...

What should I do ?

[jagabo]

What you're describing sounds like 3:2:2:2:3 pulldown. TDecimate() should work with it after TFM().

WhateverSource()
TFM()
TDecimate(cycleR=1, cycle=6)

Post a sample...

[Womb4T]

here is the sample :

[jagabo]

Mpeg2Source("sample 3.demuxed.d2v", CPU=2)
AssumeTFF()
Yadif(mode=1, order=1)
SRestore(frate=25)

[Womb4T]

Hum, ok but I don't understand the entire process xD

(seriously I hope to be as good as you are one day)

[jagabo]

Your video has field blending artifacts as well as being interlaced.

Yadif(mode=1, order=1) produces progressive frames from each field leaving a 59.94 fps results.

Some of those frames have blending artifacts (look like multiple exposures). SRestore() will preferentially throw out the blended frames, hopefully leaving only the frames with no blending.

Actually, I don't think that script is quite right. I'll have to take a closer look later...

[Womb4T]

Thanks, I begin to understand.

Indeed,

Yadif(mode=1, order=1)
SRestore(frate=25)

That s the way I manage 5i/1p patern.

So it was my first solution but it produces frame with artifacts.

(I didnt used AssumeTFF() because I don't undesrtand how it works)

[jagabo]

AssumeTFF() wasn't really necessary in that script. It tells AviSynth that the video is top field first. For example, if you wanted to use Bob() AviSynth has to know the field order. But the field order is specified in Yadif() -- order=1 means TFF.

After looking again at the video, I think the script I gave was right. Here's a quick sample after deshaking:

[Womb4T]

Well check that two screen shots (made withe the avs preview of megui). There are still jerky frame. I don't know how to handle them.

[jagabo]

You will never get perfect results with this type of conversion. Especially at the very end or very beginning of a clip.

[Womb4T]

I agree but usually I get better results with srestore. Surprsing how this one is jerky.

I am trying to find filters to improve a little bit the result.

[jagabo]

Aside from the repeated frames at the start I didn't see much jerkiness in the sample clip after that script. The jerky camera work and resulting motion blur make it hard to say for sure. Are you talking about the longer source? Maybe you should post a sample from one of the sections that's jerky after conversion.

[Womb4T]

back !

Finally I used TempMCGaus and it gave me a more acceptable result than yadif (at the great cost of speed). Thx to all for your help !

[jagabo]

QTGMC is a newer mod with some faster presets.
http://forum.doom9.org/showthread.php?t=156028

[Womb4T]

Hum nice one, gonna try it right after I fix the problem I meet actually.