Help with IVTC

[Artanis]

Im trying to rip the R2 DVD of honey & clover, but I cant manage to IVTC it properly.

I've tried tfm, IT, Decomb and SangNom, and all of them create some artifacts(I may be using wrong settings).

http://rapidshare.de/files/10360392/VideoFile-cut.m2v.html(38Mb)

Can someone give me a hand with this?

PD: Sorry for my poor english

[foxyshadis]

mpeg2source("videofile-cut.d2v").tfm().tdecimate()

Works great for me. Didn't see any artifacts at all (other than source's low bitrate, which causes a few stray combs in noisy areas). What do you get when you run that?

[Artanis]

I was refering to this:
http://img362.imageshack.us/img362/2442/6925wc.th.png (http://img362.imageshack.us/my.php?image=6925wc.png)
(frame 692 with that script)

There are a few frames like that(in the DVD, not in that sample), almost no visible in real time playback, but it annoys me to know that they are there.

[Mug Funky]

that's a problem of the original encode not having auto pulldown detection, i think. meaning it's just encoded as interlaced instead of interlaced on the interlaced bits, progressive on the 3:2 bits (which this part probably is). so in high motion it'll give interlaced artefacts...

looks like you've got a b-frame there, too. probably the best you can do with it is leave it or run a spatial blur after IVTC (even an adaptive one will blur sharp edges too). maybe a good solution is using blendbob and tdeint - with the right settings you'll blur away the stuff like that, but keep the real IVTC'able stuff. but that may not prove optimal on other frames.

[edit]

i often wonder why so many (NTSC) anime DVDs are encoded this way. the IVTC on a hardware encoder is pretty much foolproof, as it can fall back on 60i if it's unsure of a match. perhaps it's just habit to leave the IVTC option off "just in case". ADV are guilty of this almost all the time, but geneon do it quite well. too bad PAL land doesn't have the luxury...

[foxyshadis]

That was one of the exact frames I noticed, but did you take a look at the separatefields() first? (frames 1728&1729.) The original is trashed on that frame. There's not much you can do when the source has such nasty artifacts... unless it was a rip error, not original encoding error.

Throw some post-processing after it, which is a good idea for low-bitrate anime anyway. My first inclination is fft3dfilter & degrainmedian - which are good for combing and ringing, without purging details - unless you want to try the anime-specific filters in mipsmooth, tbilateral, or msmooth for stronger removal. Add one of Dideé's sharpen filters and you have your very own remastered anime! :p

Here's what I used to good effect:

fft3dfilter(plane=4,sigma=4,sigma2=1.5,sigma3=0,sigma4=0)
degrainmedian(limitY=2,limitUV=4,mode=3)
LimitedSharpenFaster()

[Artanis]

I guess that I will end letting it be(or deleting/replacing the problematic frames with freezeframe() ), thanks a lot for the help.

@foxyshadis: I'm using vaguedenoiser + this* (http://forum.doom9.org/showpost.php?p=742814&postcount=272)(not that strong, and I know that it may not be suited for this, but I tried it and liked the results) + fastlinedarken.

I dont know if it's the best option, it leads to something like this
http://img361.imageshack.us/img361/5423/clover011ho.th.png (http://img361.imageshack.us/my.php?image=clover011ho.png)

*resized to 1/2 and used 0.20 in MergeLuma(), It's fast and I love the sharpness of the result.

[foxyshadis]

It looks good and it's fast, that's all that matters. ;)