Berserk Trailer IVTC Problem

[Shinji-kun]

Sample (http://www.mediafire.com/?ionx747yy9ch4k0)

I am trying to restore a Berserk trailer i ripped from the R1 Giant Robo set. The main problem I am having is with the IVTC decimation. The problem is most apparent around 20 seconds of the video. You can see that there is an interlaced frame every 6-7 frames rather than every 5. This is apparent in that there is an "after image" of the cape ever 6-7 frames. Since the frequency of these interlaced frames is inconsistent, changing the M and N amounts via CycleR and Cycle hasn't helped.

What is the cause of these interlaced frames and what parameters or extra filters could take care of the issue?

Script:

DGDecode_mpeg2source("C:\Berserk Trailer.d2v")
TFM(order=-1,field=-1,mode=1,d2v="c:\berserk trailer.d2v",slow=2,pp=5).TDecimate(mode=2,noblend=true)
source = last
Frfun7()
filtered = last
ContraSharpening(filtered,source)
gradfunkmirror()
AAA(Xshrp=20,Yshrp=20)

[Hagbard23]

Okay - it is not the smartest way, but it should work as you want it to ... try it out. I use it, when TFM and TDecimate are not able to inverse telecine correctly.

http://forum.doom9.org/showthread.php?t=60031

But maybe someone has a even better idea...

You uploaded the sample in another thread some weeks ago, i tried it with smartdecimate() and it worked quite good.

[Shinji-kun]

Tried Smartdecimate() and there are still combing artifacts/interlaced frames. What parameters did you use?

[Didée]

SmartDecimate is always a close call on parametrisation - a little too little gives combing, a little too much gives false duplicates, ...

Point is that the section in question is not a standard telecine. Instead of field pulldown, there are blended fields inserted. I don't think that any of the standard IVTC-filters is up to handle that. They don't look for blended fields, and since the blendfields are "unique", they are seen as valid content.

The fieldblended sections could be handled by Srestore, but chances are you have to isolate those sections and process them separately. Maybe you can try to abuse Srestore to do all the IVTC ... I never tried that, perhaps it works out.

Yet another possibility is to have a deeper look into the "AnimeIVTC" function ... it has lots of mysterious processing modes that I don't understand. (Never working with Anime, you don't know all those "impossible" perversions that you can find only in Anime.)

[Hagbard23]

Hmm - i used no special parameters, but it was just a quick shot - i would suggest, that you tweak around a littlebit by your own. I would fine-tune the following parameters (see here: http://kevin.atkinson.dhs.org/tel/SmartDecimate.html):

1. use DGBob instead of avisynth-bob:

b=dgbob(order=[set the right field order here*]) [*see documentiation of DGBOB() for details]
Smartdecimate(Bob=b)

2. tweak "tel" parameter:
i would lower the default value of 0.5 to -lets say- 0.3 or 0.2 if i had difficult video

3. tweak "noise" parameter
hmm - this is definitely dependant to your video - i would try a littlebit around and see what happens...maybe you rise it a littlebit.

4. add post-deinterlacer for "cleaning up" the leftovers...
I would say, the few left interlaced frames are not really worth getting a headache of it. So i would clean them away with a simple deinterlace filter.
Hmm - i do not know if you can use triticals NNEDI Plugins for that purpose. If i had to post-deinterlace i would use something like YADIF (http://avisynth.org.ru/yadif/yadif.html) or tomsmocomp or something similar. There a various deinterlace-plugins, and i think YADIF is a good choice overall.

After that - i would do all the post-processing - i definitely would use a decent sharpener for your video. Warpsharp or LSF or LSFMOD should to the job rather good, but especially LSFMOD is quite slow and should be tweaked with intelligence, although the default settings may fit to your purposes - i would try it out.

Maybe you want to denoise a bit - for that purpose FFT3dGPU may fit - but there are various other plugins to play with.

I would NOT use a deband / Gradient filter - i think your video doesn't need that.

You can replace the Antialiasing function -at least a part of that- via LaTo's LSFMOD if you turn supersampling on - but be aware: Supersampling is "Super-Slow". Keep in mind, that LSF and LSFMOD work rather good with Multithreading - but you have to have a "hacked" version of AviSynth for that; either SeT's Avisynth MT or Joshy&SeT's AviSynth x64 MT. See here for details: http://avisynth.org/mediawiki/MT

At least:
That would be my way to do the job - other (maybe more experienced users) would go other ways, i don't know yet. But i DO know, that my way would lead to -lets say- satisfying results.

If you have further questions, just ask...

[Shinji-kun]

Tried encoding to x264 with this

DGDecode_mpeg2source("C:\Users\dvd rip\Berserk Trailer.d2v")
LoadPlugin("C:\program files (x86)\avisynth 2.5\plugins\removegrain.dll")
LoadPlugin("C:\program files (x86)\avisynth 2.5\plugins\yadifmod.dll")
TComb(mode=0)
AnimeIVTC(mode=2,aa=0,omode=1,credconv="blend")
Temporaldegrain()
AAA()
gradfunkmirror(strength=1.5)

When I got to 99% of the second pass MEGUI gave an error saying "AAA() it looks like we've run out of memory folks" or something similar, and then the job was stopped without encoding the last ~30 sec. of the video. When I watched this video it still had aliasing and grain and some blended (after image) frames.

A few questions.

1. I've narrowed the AnimeIVTC mode down to 2 or 3. the problem is though that there are still blended frames. Does anyone know whether the footage I posted in this thread is Field blended, or Double Hard telecine?

2. What is the error with AAA() and how can I fix it?