Jumpy Noise on Anime, Maybe Bad IVTC

[WSFrazier]

I am new to anime encoding and would love to learn and I am starting slowly. My question is the anime I am playing around with always turns out jumpy. By that I mean the lines mostly. Seems there is always moving noise too. Just for the sake of it I tried oversmooth it to hell and it is still moving. I watched a couple encodes done by a few groups and noticed that all their encodes turn out perfectly stable. I am curious as to how they keep the picture so steady. I guess that is one of their secrets.

I am thinking it may be my IVTC or deinterlace. Right now I am using TMPG to Auto IVTV and Auto Deinterlace. I then save the TPR file and import it into my AVS script. Is this a bad way to do it? and could that be reason?

[Chainmax]

It's been a very long time since I last used TMPG, but I'd bet that their IVTC algorithm is outdated by now. The only reason I would use it is as a base for manual IVTC. Otherwise, you are probably better off with TIVTC (http://forum.doom9.org/showthread.php?t=82264&page=22&pp=20). Besides, I think that TMPG needs RGB, so there's an unnecessary and costly colorspace conversion that takes place (DVD is YV12 and a capture is usually YUY2).

This moving noise you're referring to might be DotCrawl, for which I'd recommend to use DeDot() (http://nullinfo.s21.xrea.com/#DeDot_YV12) before IVTCing or Deinterlacing.

You shouldn't deinterlace if you are doing IVTC, as IVTC filters usually have builtin postprocessors to eliminate whatever leftover combing might remain from the field matching. In any case, you seem to not be familiar with the concepts behind IVTC and deinterlacing. I highly recommend that you read the readmes and tutorials of the Decomb package (http://neuron2.net/decomb/decombnew.html). They are excellently written and are about the best startup guide you'll find. Also read this (http://www.doom9.org/ivtc-tut.htm). Those will help you out a great deal. If you have any further doubts, don't hesitate to ask.


And for future references, try to post a small vob sample so that we can help you better. You can use ChopperXP (http://www.inchwest.com/chopperxp.html) for that. If you're capturing, just cut out a small sample with VirtualDubMod.


P.S: welcome to Doom9 :).

[WSFrazier]

Here is a sample vob.

http://www.icon-clan.com/Sample.vob

Here is a little clip that I did with xvid, I know it is oversmoothed but like I said I was messing around.

http://www.icon-clan.com/Sample.avi

If you look at the part where the man with the helmet is falling (beginning) the lines are jumpy. From the source it is rainbowed and when that rainbow is removed the line goes with it. At least I think, new at encoding anime :D Is there something that can be done to stabilize this? Maybe something that can fill and make the lines solid? Also any more suggestions on what I can do to improve would be great.

[Chainmax]

I'll see what can be done.

[Guest]

Use DGIndex with Force Film. That's all the IVTC you need. It gives the original clean progressive frames. The remaining artifacts are in the source. Your main problem is deinterlacing when you don't have to.

[Chainmax]

I see many people that always try to use Force FILM if they can, is it really any better than DeComb or TIVTC? After all, isn't it supposed to only work flawlessly if the stream is 100% FILM?

[Guest]

I see many people that always try to use Force FILM if they can, is it really any better than DeComb or TIVTC? After all, isn't it supposed to only work flawlessly if the stream is 100% FILM? Yes, of course, it is only viable (without postprocessing) if the stream is pure flagged 3:2 pulldown. But there are many, many sources that are just that. In such cases, Force Film is greatly preferable because it requires no additional filtering, which can be slow, and totally avoids the possibility of false field matches and consequent postprocessing artifacts. Also, it is often useful when the material is only slightly non-pure 3:2 pulldown.

[Chainmax]

Why don't DeComb or TIVTC have such a mode of operation then?

[tritical]

Why don't DeComb or TIVTC have such a mode of operation then?
TIVTC has since the end of july... See the possible settings for the flags parameter in combination with the d2v parameter. By default (flags = 4), if the d2v parameter is specified tfm will use the d2v for field matching in film sections but will check the matched frame for being combed (if it is combed it uses its own matching routine for that frame). If you set flags=1 then it will do the same thing, except that matches from the d2v will not be checked for being combed and will always be used as is. Even with flags=1, using TIVTC will be slower than using forcefilm, but it will still be quite fast (about 450fps for the combination of tfm/tdecimate on my 1.6Ghz laptop) and provides extra safety on clips that aren't 100% film.

[Chainmax]

I stand corrected.

[WSFrazier]

Is there anything I can use on those lines to make them more solid or less jumpy? I tried FastLineDarken but that does just what it says and makes the jumpiness stand out more.

[DryFire]

depan?

[Anonymouses]

Is there anything I can use on those lines to make them more solid or less jumpy? I tried FastLineDarken but that does just what it says and makes the jumpiness stand out more.

I'm not sure what you mean by jumpy lines. If you are talking about how the lines look slightly different between frames thats just due to the way it was animated and there's not much you can do to fix it. Such inconsistency is what you get from cel-drawn anime.

[Chainmax]

Wait a minute, that vob sample has a 23.976 framerate. Please provide a completely unprocessed sample.

[Anonymouses]

Wait a minute, that vob sample has a 23.976 framerate. Please provide a completely unprocessed sample.

Huh? I get 29.970 from DGIndex.

http://img441.imageshack.us/img441/2919/thing7ps.png

[Dreassica]

looks like someone forgot to turn off Force Film in dgindex :P

[WSFrazier]

I used ChopperXP to cut it straight from the DVD.

[Chainmax]

I merely opened the sample straight on VDubMod, no kind of postprocessing was enabled on ffdshow either.