Anime encoding problems...

[arch_angel16]

Love Hina DVD, GKnot 0.27, Divx 5.02Pro. Tried to encode an episode, used IVTC in dubber (not force film). Worked (mostly), but there are some interlaced frames left...just enough to be annoying, esp since it's mostly the mouth moving that gets the interlace artifacts. Question is, would deinterlacing after the IVTC work? or a more aggressive IVTC???... how do i do these things?

[jggimi]

If you have not already seen it, take a look at www.doom9.org/ivtc-tut.htm

[arch_angel16]

already read it, didn't help much.

[manono]

Hi-

You've got lots of choices, and when working with anime, you have to understand and be prepared to experiment with and use the different IVTCs, and all their options. First, get the DecombLegacy (http://shelob.mordor.net/dgraft/decomb/decomb405legacy.zip). Then rename the DecombLegacy.dll as Decomb.dll and throw it into the GKnot folder. Then try:

Telecide(Guide=1,Agg=True)
Decimate(5)

Open the .avs in VDub, scroll to some of the places where you have the problems and see if that fixes it. Or put the deinterlacer on more strongly:

Telecide(Guide=1,Agg=True,threshold=5,dthreshold=5)
Decimate(5)

Or if you have only a few problem places, use an override file to fix the messed up frames individually (see the DecombHelp.html). Or open the IVTC Plugin in the .avs and use this instead:

IVTC(70,10,25)
FieldDeinterlace()

Or use the IT Plugin (http://members.tripod.co.jp/thejam79/IT_0051.zip) to IVTC instead.

Question is, would deinterlacing after the IVTC work? or a more aggressive IVTC???...

Decomb has a deinterlacer in it by default. But yes, a more agressive setting for the deinterlacer, or more aggressive settings for the IVTC, or if worse comes to worst, a different IVTC might solve the problem. Read and understand the DecombHelp.html

[jggimi]

All I know of "Love Hina" was that it was made-for-TV, according to http://us.imdb.com/Title?0279570. It may not be Telecined, but instead, may be interlaced. This would be the case if it was shot on video rather than a transfer from film. Did you examine the content, frame-by-frame, prior to encoding, to determine if it was interlaced or Telecined?Just to be clear on the IVTC process you used, did you select IVTC from within Gknot, or did you make any manual changes to the .avs script? The reason I ask is that if your IVTC (or deinterlace) was done manually in the script, if the commands follow after resizing, interlaced artifacts will remain.

[arch_angel16]

i used IVTC within gknot at the 2nd last stage before encoding (when the dvd2avi has the menu for ivtc, deinterlace, subtitles, comp check, etc)

When i looked at it frame by frame, the general impression i got was 1 or 2 interlaced frames for every 4 or 5 noninterlaced frames. So i figured it was telecined. Thing is, sometimes the anime behaves as if it were interlaced, esp while there are scenes with heavy dialogue.

Mmkay, gonna try those, but where do I insert these text commands into the settings?

[Guest]

Most "mouth" problems are solved by proper use of pattern guidance and/or the mm option. One thing overlooked by many is the proper use of the mm option, as it can substantially reduce false matches. I need to explain it better. Basically, you run first with mm=0 and observe with show. If you see primarily good c & n matches and only sporadic p, use mm=2. If you see primarily good p & c matches and sporadic n, use mm=1. In my experience, it is rare that a clip validly combines n and p matches; in such cases, use mm=0.

I'm not sure I like agg. :)

[arch_angel16]

Telecide(Guide=1,Agg=True)
Decimate(5)

Thanks, that worked ^_^.

And anime IS the bastard child of video encoding...LOL...

So, if it worked this way, that does mean Love Hina is telecined, and not interlaced? I've noticed this in newer anime releases, like Hellsing.

[Guest]

Originally posted by arch_angel16
Telecide(Guide=1,Agg=True)
Decimate(5)

Thanks, that worked ^_^.

And anime IS the bastard child of video encoding...LOL...

So, if it worked this way, that does mean Love Hina is telecined, and not interlaced? I've noticed this in newer anime releases, like Hellsing. The LH episodes I have seen are cleanly 3:2 telecined. Hellsing is not clean. :(

EDIT: Hello manono! Long time no chat. Hope you are well.

[arch_angel16]

I had no problems with Hellsing whatsoever, and I've watched my divx rips of them quite often. Amazing compression too, <200MB per ep at 100% of the bit/frame deal

[jggimi]

...that does mean Love Hina is telecined, and not interlaced?

It is telecined, because...

...the general impression i got was 1 or 2 interlaced frames for every 4 or 5 noninterlaced frames. So i figured it was telecined...

You figured correctly.

[mazzo]

This is a NTSC problem, isn't it? I had a similar problem with the Simpsons (PAL), but got descent result with soft bicubic resize and field deinterlace.

[Guest]

Originally posted by mazzo
This is a NTSC problem, isn't it? I had a similar problem with the Simpsons (PAL), but got descent result with soft bicubic resize and field deinterlace. What does your "This" at the beginning of the sentence refer to?

[mazzo]

Inverse Telecine

[Guest]

Inverting 3:2 pulldown is an NTSC issue, yes. But even in the PAL world there are cases where IVTC is required. For example, a 24fps film may be telecined to 25fps by adding 2 fields per 48. Decomb can undo it with this script:

Telecide()
Decimate(cycle=25)

[Justinus]

I heard that you could use GreadHMA for PAL DVD with Telecide as well, couldn't you? It says it was designed to do 2:2 pulldown, specifically for PAL Anime.

Anyway, I'm currently working on Love Hina Again (NTSC) and still couldn't get rid off all interlaced artifacts. I'll try your code and IT tonight.

One question...if there's a short range of video clip looks like interlaced video but couldn't be gotten rid of even tried using IVTC or FieldDeinterlace, what could it be? I even tried to overide Decomb but it couldn't do anything.

[jggimi]

...video clip looks like interlaced video but couldn't be gotten rid of... You can check to see if the artifacts are in the source using SeparateFields().SelectEven() or .SelectOdd().

If so, there is little or nothing that can be done to remove the artifacts if you want to retain reasonable resolution.

It may be possible to remove them using something like VerticalReduceBy2() after selecting a single field, but you'd end up with 1/4 of the original resolution at that point.

For example, 720x480 after SeparateFields().Select<field>() would be 720x240, then doing an additional VerticalReduceBy2() would give you a resolution of 720x120. If your original input stream was 720x576, then your output would be 720x144. Either way, I doubt you'd be happy with it.

[Justinus]

@jggimi
Thank for your invaluable advice :D

I'll try it tonight and see what happen. In any case, could you explain to me why I have to do like this. Is there anything wrong with that portion of the source? :confused:

BTW I'm sure that I wouldn't be happy with the movie with the height of 144 or 120 pixels for sure (looks like I'm watching the movie from a handheld device). Anyway, I'll do what you told me first.

[manono]

Hi-

I think using SeparateFields().SelectEven() followed by VerticalReduceby2() might work if the interlacing is within each field (pretty rare, but I guess it happens). And afterwards you'd resize to 512x384 or whatever you use. But as jggimi says, it would probably look pretty nasty.

But if I'm guessing correctly, that's not your problem, since you're working with NTSC material (interlaced fields are found in PAL anime material usually).

Assuming yours is a case where some of it just isn't being deinterlaced fully, you might try:

Telecide(Guide=1,Agg=true).Decimate(5).

And if that doesn't clean it up, then you can make the deinterlacer stronger with something like this:

Telecide(Guide=1,threshold=5,dthreshold=5).Decimate(5).

And if that still doesn't clean it up, then you can try:

Telecide(Guide=1,Post=false).Decimate(5)
BlendFields()

And that should get rid of all interlacing (but use it only as a last resort). And you can also take neuron2's advice earlier in the thread and use the mm option (see the DecombHelp.html for more information). Did you read the other advice earlier in the thread about how to handle Love Hina? And to use the Agg=True setting, you'll need the DecombLegacy.dll linked to earlier. It's more recent than the version of Decomb included with GKnot.

[Justinus]

@manono
I haven't tried all those command lines mentioned previously yet because I got the problem with VirtualDub last night.

Anyway, I'm using AviSynth 2.5.1 beta at the moment (kinda interested in YV12 colour space) so that I use Decomb 4.06 (BTW I didn't use GKnot).

And...yes I've read all previous posts through. I've tried using agg=true and also change both threshold and dthreshold into smaller value. It helped a lot for blinking eyes and moving mouths but it didn't show any slightly effect with that portion. I'll try to put some more combination based on the previous posts again, including yours.

What I'm guessing now is that portion might be badly interlaced. Don't worry...I'm not giving up now as I'm just having fun deinterlacing NTSC source (I'm living in PAL country so I don't normally have a chance to have any trouble with deinterlacing process) :D

[Guest]

Originally posted by manono
I think using SeparateFields().SelectEven() followed by VerticalReduceby2() might work if the interlacing is within each field (pretty rare, but I guess it happens). What do you mean by "interlacing is within each field"? Thank you.

I'd be more inclined to suspect blended fields.

[manono]

Hi neuron2-

I was reading a post here just the other day about bad PAL anime where the poster mentioned that after doing a SeparateFields() he saw interlacing in the fields. The thing is though, I have no idea where I read it. I'll try and find it. I don't know if it's technically possible for that to happen. I was just passing it on.

All I know is what I read in the papers-Will Rogers.

[Guest]

If it is YV12 that can be an artifact and not interlacing. I'd be interested in seeing the source clip.

By definition, a field cannot be interlaced.

[manono]

Justinus asked me to attach this zip file showing his problem. It's a small piece (317 KB) of the .avi, together with a .txt file containing his settings.

[Guest]

Attachments aren't working. You have to host it somewhere and post an URL.

[manono]

Sorry, I just noticed that. I should have checked it right after I put it there to make sure, but it looked as if it were OK. I'm getting a .vob from Justinus now, and will contact you when it's ready.

[Kenshin5]

Hi guys i tried this method:
Telecide(Guide=1,Agg=True)
Decimate(5)

And i get the error: (Telecide does not have a named argument "Agg")
what's wrong i loaded up the Decom521.dll i dunno what else to do.

[manono]

Hi-

The Agg setting went out after Decomb 410. You can still find it at neuron2's site, I think.