IVTC Anime

[pwnsweet]

Hi guys hows it going. I'm looking at taking my R1 Death Note DVD rips and encoding them for use and storage on my computer. I didn't expect this to be a particularly difficult task however it would seem that I was very wrong about this. I have spent numerous days on this single problem and it is getting quite frustrating. I am going to try and detail everything I have done/tried so that there is no misunderstanding of facts.

Overview of what I have done:

1. I Re-authored the DVD discs with DVD Shrink (http://www.videohelp.com/tools/DVD-Shrink) v3.2 and trimmed the main video file to keep the VOB files for each individual episode (without the Intro/Outro). I did this with no compression so it was a direct copy. I have uploaded a short segment of the VOB file via mediafire here (http://www.mediafire.com/?xd2jc476kyganed).

2. I then proceeded to demux the video/audio tracks using DGIndex (http://www.videohelp.com/tools/DGMPGDec) v1.5.8 and under the video tab I set the Field Operation as "Honor Pulldown Flags".

3. I then imported the resulting .m2v video file into TMPGEnc Video Mastering Works (http://www.videohelp.com/tools/TMPGEnc-Xpress) v5.0.6.38 and applied my desired color/sharpening filters. I plan on encoding each episode using TMPG (http://www.videohelp.com/tools/TMPGEnc).

4. The m2v file is reported as having an interlaced framerate of 29.97 fps and after inspecting the fields I can conclude that the the video was originally filmed at 23.976 fps and has been Telecined up to 29.97 fps (no surprise, this is what you would expect from an anime that aired on TV). The interlacing order is Top Field First (TFF).

Just to be clear, what I see on the video is this pattern:

AA
BB
BC
CD
DD
EE
FF
FG
GH
HH

The Problem

The problem lies in encoding the video to 23.976 without any judder/artifacts. TMPG has its own de-interlacing filter under the filter tab and I have tried every combination of settings under the de-interlacing filter (theres 50 combinations) and NONE of them have a non-judder image throughout the entire episode. Some settings would result in perfect 23.976 progressive frames in certain scenes but other scenes would be a total mess. This is the first time I have had this problem in TMPG and so this led me to believe that the video is the problem and not TMPG.

After closer inspection of the video file I found that after every scene change (by scene change I mean a 'camera cut') the pattern of the 3:2 pulldown is broken! The new scene still has the same 3:2 pattern as outlined above however it is not consistent with the previous scene. For example this might happen:

AA
BB
BC
CD
(scene change)
DE
EF
FF
GG
HH

You can see here that the 3:2 pattern is broken, this happens at every single scene change. This is why a simple Inverse Pulldown filter will not work, the pattern is inconsistent. I have just started trying to learn Avisynth (http://www.videohelp.com/tools/Avisynth) and have attempted to use some of the standard IVTC filters without any success. It does look like I will need to use some sort of avs script to get this working though...


And so I turn to you, my knowledgeable internet brethren, in the hope that a solution to this problem might be found.

[mp3dom]

Majority of anime are made in that way. They animate the scenes at 23.976, then they apply the pulldown to 29.97 and then they start to do the real editing. They cut scenes when the scene itself is already at 29.97 so they broke pulldown patterns, also they do real postproduction at 60i (for example rolling credits, fadeout, dissolvences between different scenes, cross dissolvence with scenes with different pulldown patterns and so on). They work in this way because the final output is for television so they don't care about these kind of problems. Your best bet is to search here for AnimeIVTC which should be able to handle different cases (regular pulldown, 60i patterns, interlaced fades and so on). If you want a more 'precise' control you can try to use YATTA but I warn you that this is very time consuming and can be frustrating for 37 episodes!

[pwnsweet]

I have looked at AnimeIVTC, it slowed my computer down to a complete crawl. I'll have to try out YATTA. So would this be under the "Double Hard Telecined" category or does this type of problem come under a different name?

Edit: YATTA appears to be quite confusing...

[TheRyuu]

Your sample doesn't have anything out of the ordinary. A simple IVTC script will work fine on it:
Mpeg2Dec3_Mpeg2Source("X:\path\to\deathnoter1.d2v")
tfm()
vinverse()
tdecimate(mode=1)
And yes it will take into account the pattern changes between scenes (hence why we use tfm/tdecimate instead of a simple force film from within dvd2avi/dgindex). If you don't understand yatta you should probably avoid using it as it will most likely do more harm than good if you have no idea what you're doing. If you are having additional problems that perhaps your sample doesn't suggest then upload another sample which displays the issue.

Vinverse is for the residual combing left after matching (R1 bullshit I guess).

[Guest]

(hence why we use tfm/tdecimate instead of a simple force film from within dvd2avi/dgindex) The sample is hard telecined, so force film mode is out of the question, regardless of any pattern phase changes.