Strange input; ideas?

[Dada_]

Hi there. I've been working on a translation for an old anime for some time now, and with the first episode finished it's time to learn how to actually make some pretty encodes.

I've done some reading and tried a bunch of different things with MEncoder and AviSynth, but I haven't really had any luck so far. My input is quite strange.

Here's my input: http://tmp.thingmajig.org/ajk_input.avi (30 MB)

It's ripped from a PAL DVD (I'm actually subbing it from Dutch to English, so unfortunately there's no other option), the FPS being 25.

I tried using AnimeIVTC(), but with every frame being interlaced, I wonder if I should. I used mode=5, but it didn't work out too well. (Some scenes were perfect, others showed the ghosting effect.)

So what I'm thinking now is maybe I should just bob everything and make it 50 FPS.

Does anyone have a recommendation for me? And is this input as "difficult" as I think it is?

Thanks!

[Leak]

Originally Posted by Dada_

Here's my input: http://tmp.thingmajig.org/ajk_input.avi (30 MB)

It's ripped from a PAL DVD (I'm actually subbing it from Dutch to English, so unfortunately there's no other option), the FPS being 25.

You really should post a small piece of the original VOB file (you can cut it out using DGIndex, for instance), as only that allows analyzing your source in detail...

That being said - I'd either go with NNEDI(field=3) or encoding it interlaced and bobbing on playback (if neccessary) because with all the scratches, the wobbling between the odd and even fields and everything else you probably won't have much chance of field matching and VFRing Alfred and friends anyway... and it's not telecined either, so there's no need to try and IVTC it.

np: The Orb - S.A.L.T. (Snow Mix) (Orblivion Versions)

[Dada_]

Originally Posted by Leak

You really should post a small piece of the original VOB file (you can cut it out using DGIndex, for instance), as only that allows analyzing your source in detail...

I just got DGIndex. Here's an m2v file: http://tmp.thingmajig.org/VTS_01_1.demuxed.m2v (36 MB). I'm not sure if this is what you meant, but I couldn't figure out how to get an actual VOB crop.

Originally Posted by Leak

That being said - I'd either go with NNEDI(field=3) or encoding it interlaced and bobbing on playback (if neccessary) because with all the scratches, the wobbling between the odd and even fields and everything else you probably won't have much chance of field matching and VFRing Alfred and friends anyway... and it's not telecined either, so there's no need to try and IVTC it.

One question about this: isn't it so that AnimeIVTC() has more functionality besides inverse telecine, so that it would probably be a good idea to use it anyway? Maybe as some sort of post-processing pass?

Also, I'm gonna go try NNEDI right now.

EDIT: well, NNEDI(field=3) didn't have the desired effect. Now the image bops up and down a pixel or so on every frame. It looks very strange. Should I upload an example of the output?

EDIT: link above finished uploading, so it can be downloaded now.

[Guest]

It's just phase-shifted PAL. Use a field matcher like Telecide() or TFM().

[Dada_]

Just tried using Telecide(). It doesn't look too appealing, in my opinion: http://tmp.thingmajig.org/ajk_telecide_output.png

But maybe this is the best possible interlacing with my input, and I just need to use some AA or other filtering to take care of this?

[Guest]

Your link is dead.

What was your script and which frame number of your sample are you talking about?

[Leak]

Originally Posted by neuron2

Your link is dead.

That's strange - it works fine for me...

@Dada_: that up- and down-bobbing is right there in your source, probably introduced when they turned their progressive master into something interlaced for TV. That's also why the picture still looks slightly combed after Telecide() or TFM(). (Also, I had to lower MI to 40 with TFM or it would sometimes match the wrong fields...)

If you want to get rid of any combing, use NNEDI(field=1) instead. I don't think you'll lose much details that way with this source...

np: The Orb - Secrets (Orblivion Remastered)

[Dada_]

Originally Posted by neuron2

Your link is dead.

What was your script and which frame number of your sample are you talking about?

It might have still been uploading. It's up now. Basically, it was just vanilla Telecide():

AviSource("c:\avisynth\input.avi")
loadPlugin("C:\Program Files\AviSynth 2.5\plugins\Decomb.dll")
AssumeTFF()
Telecide(guide=2,post=2,vthresh=1,show=true)

I figured I'd set the vthresh to 1 because every single frame is interlaced, so there's no need to try and detect any progressive ones.

Originally Posted by Leak

@Dada_: that up- and down-bobbing is right there in your source, probably introduced when they turned their progressive master into something interlaced for TV. That's also why the picture still looks slightly combed after Telecide() or TFM(). (Also, I had to lower MI to 40 with TFM or it would sometimes match the wrong fields...)

If you want to get rid of any combing, use NNEDI(field=1) instead. I don't think you'll lose much details that way with this source...

np: The Orb - Secrets (Orblivion Remastered)

Thanks, I'm gonna go try NNEDI(field=1) right now, will edit later to see how it worked out. I guess that if only I can get rid of the combing while still preserving details as much as possible and without introducing ghosting, I'll be happy for now. Then I'll just have to see what I can do with other filters to compensate for the rest.

EDIT: alright, this looks quite a bit better: http://tmp.thingmajig.org/ajk_output_nnedi1.mkv (31 MB). It's still uploading but it should be done a short while after I make this edit. The up-and-down bopping is gone, and the input is a bit less sharp, but there's no ghosting and the image looks quite good now. I'm wondering: is this the best I can do as far as quality goes? The edges still look a little iffy and some sharpening might be nice. I'm probably going to crop it a little bit to get rid of the black bars. I think maybe some anti-aliasing will help as well. And does anyone know if it's possible to compensate for the fact that the entire screen tends to slowly move around a little bit?

[egrimisu]

What is this anime caled?

[TheRyuu]

Mpeg2Dec3_MPEG2Source("C:\Documents and Settings\Dragon\My Documents\My Videos\phase shifted pal.d2v")
SetPlanarLegacyAlignment(true)

assumetff()
tfm(slow=2)

That worked just fine for me.

[Dada_]

Originally Posted by egrimisu

What is this anime caled?

Alfred J. Kwak: http://en.wikipedia.org/wiki/Alfred_J._Kwak

Originally Posted by Dragon5152

That worked just fine for me.

Many thanks, I'm going to try this when I get home from work tonight!

[2Bdecided]

It is field shifted, but even if you correct that, there's another problem.

The film scanner was faulty - that's why the two fields don't match. There's no point using something like telecide which tries to match them - it's not going to work - they consist of the same frame of film, but that physical piece of film moved in the gate between fields, giving images that cannot be matched.

The only sensible approach (unless you want to correct for the film gate warp!!!) is to dump one of the two fields, and to reconstruct the missing lines using EEDI2 or NNEDI.


The strange thing is that MPEG2source hangs with this clip for me - I've never had this problem before. Converting to AVI in VirtualDub leaves ghosted colours that might not be in the original - if they are, that's another challenge.

Cheers,
David.

[Dada_]

Originally Posted by 2Bdecided

It is field shifted, but even if you correct that, there's another problem.

The film scanner was faulty - that's why the two fields don't match. There's no point using something like telecide which tries to match them - it's not going to work - they consist of the same frame of film, but that physical piece of film moved in the gate between fields, giving images that cannot be matched.

The only sensible approach (unless you want to correct for the film gate warp!!!) is to dump one of the two fields, and to reconstruct the missing lines using EEDI2 or NNEDI.

Hi there. Thanks very much for shedding some light on this. This is very interesting stuff to me. Does this mean that the producer was shoddy when digitizing it?

NNEDI already got me some good results. I guess I'll have to drop one of the two fields, then. Do you recommend EEDI2 or NNEDI?

Final question: bobbing the input to double the framerate is also inevitably going to cause the "wobbly" effect, right? I've tried bobbing it before and the output kept hopping up and down on every frame. But there's no way to avoid that?

(Haven't tried the last suggestion yet since I've been busy, but I will do this very soon.)

[2Bdecided]

I think the film scanner / telecine machine was faulty or misaligned. Probably ready for a service, or the scrap heap.

EEDI2 or NNEDI? Whichever looks better to you. I use EEDI2 - totally subjective decision.

No point bobbing - that keeps both fields, and you must dump one with this faulty source.

Cheers,
David.

[Nightshiver]

Wouldn't Yadif(order=1) work just fine? It did for me on the sample...