Progress with hybrid telecine DVDs

[silver_cpu]

I'm desperately hoping that this hasn't been covered completely in another thread (I'm sure it has, I'm just waiting for someone to point it out to me). After much, much reading, I've come to the conclusion that there is no current solution for the "Hybrid framerate DVD" problem that occurs occasionally. What happens is that a DVD (most often a TV show) will be captured on film (24fps), and then during editing some 29.97fps material (usually special effects and CG material) will be added in. Thus, you are left with a choice: telecine everything, reducing it to a flat 24fps and getting choppy scenes with the 29fps material, or encoding everything at 29fps, and losing quality to duplicated frames.

This makes the ripping of many anime titles (the bastard child of DVD ripping) and (the big thorn in my side) the Star Trek Next Generation DVDs difficult (at best) to rip. Is there *any* way to solve this, or is anyone working on a viable solution? Surely this isn't such a strange problem that some work-around can't be devised. Maybe DVD2AVI could be updated to recognize the changes in framerate (and oddball telecine procedures, for that matter), and create a .d2v file with appropriate flags, I'm not sure (don't know enough about that to say). I'd just like to be able to rip DVDs, and know that I didn't make compromises like these :)

Once again, I hope I'm not asking a question that's been asked many times over in the forums (I searched around to no avail for any current topic on it), but I think this is an important issue that needs to be resolved eventually. Thanks!

[cheap-red-wine]

Here are the possible solutions, in order of quality:

1. Leave it as is and play it back at 29.97fps interlaced, e.g., via a TV out, or on a standalone player.

2. Separate the film and video portions and process the film with IVTC and the video with motion-compensated frame rate reduction (e.g., with MotionPerfect). Reassemble the clip at 23.976fps.

3. Use a field decimation method to reduce the video to film rate. This is usually acceptable for natural video but leaves objectionable artifacts for anime and CG. Field decimation is hard to do without artifacts and/or significant degradation of vertical resolution.

4. Leave it at 29.97 but replace the duplicate frames with a blend of the frames before and after. Or IVTC it to 23.976 but use blending on the video. Decomb's Decimate() can do these.

It is possible to trick DVD2AVI to output the IVTC portions without pulldown and the video portions without decimation, but this produces audio desync and the wrong playing speed for either the film or video portions (depending upon the final frame rate you choose).

In practice I find it easiest to just leave such clips at 29.97 by field matching and then deinterlacing. The duplicates are objectionable only in fast pans.

[silver_cpu]

After some experimentation (I'm using the infamous STNG DVDs), I've found that encoding at 29.97fps with Tom's Mocomp filter works pretty well. It handles everything without artifacts or odd defects, the audio stays in sync, and lo and behold the video still compresses nicely. I think I'll stick with this solution, until DVD2AVI comes up with a way of handling this problem (which is becoming more common now, as CG and film are used in many many TV series now).

[cheap-red-wine]

But why destroy your good progressive frames if you don't have to?

And what makes you think there is a solution that can be implemented in DVD2AVI?

[fccHandler]

5. Encode the video at 119.88 fps. Duplicate each NTSC frame 4 times (or each field 2 times), and duplicate each FILM frame 5 times.

[silver_cpu]

cheap-red-wine:

But why destroy your good progressive frames if you don't have to?

What do you mean? I'm afraid I don't understand why encoding at 29.97fps would do anything to the progressive frames (besides duplicate them). If I'm not mistaken, the argument goes the other way around (encoding at 24fps makes the interlaced material jerky). Remember, I'm encoding the hybrid Star Trek DVDs, which have film-rate material (most anything that isn't CG), but with 30fps CG material. Perhaps my thinking is wrong, but I believe that encoding at 30fps saves the CG from stuttering, but suffers from poorer quality since there are more frames to encode than before(since there are more frames to encode, there are fewer bits to go around). This is something of a half-solution, though, as you still are forced to encode at a fixed framerate, whereas the DVD seems to not be limited in this fashion (once again I might be wrong there). If, however, I'm wrong and encoding at the faster framerate harms the quality of the progressive frames (above the aforementioned bitrate problem), please let me know. I didn't notice anything wrong with what I've encoded thus far, but I'm not an authority on the subject.

And what makes you think there is a solution that can be implemented in DVD2AVI?

Sorry if I poked at DVD2AVI, I figured the stats generated at the demuxing stage would be most appropriate for use in automating the re-encoding. I wouldn't expect VDub to be able to do much, if the D2V file stated a consistent framerate (maybe I'm wrong here, I'm afraid I'm unfamiliar with the technological details in DVD ripping).

fccHandler:

5. Encode the video at 119.88 fps. Duplicate each NTSC frame 4 times (or each field 2 times), and duplicate each FILM frame 5 times.

I'm assuming that this came from a FAQ or readme that I must have missed (I do try and read them first). In any case, I have no clue what it means. Is this a joke? And if not, why in the world would you want to duplicate all that data? Wouldn't it seriously hurt the visual quality of the encoding?

[fccHandler]

Originally posted by silver_cpu
I'm assuming that this came from a FAQ or readme that I must have missed (I do try and read them first). In any case, I have no clue what it means. Is this a joke? And if not, why in the world would you want to duplicate all that data? Wouldn't it seriously hurt the visual quality of the encoding?
No, it came from an experiment I posted to this thread (http://forum.doom9.org/showthread.php?s=&threadid=43362) a while back. The link there is dead, but I think I still have the file if you want to see it.

The data need not be duplicated. The duplicate frames in the AVI file could be written with zero size (like dropped frames), but you would need a special tool to do that, and I don't know of one. What I found out was that MPEG-4 codecs love the duplicates (especially if they coincide with delta frames), they compress to practically nothing, and the resulting size is not much larger than any normal 29.97 movie.

[silver_cpu]

Thank you, fccHandler, for your post. One last question, and then I think I can be satisfied: when DVD2AVI demuxes the video stream from the VOB file, and you select a framerate of 29.97fps, does it automatically duplicate frames in the 24fps material? If so, then I can be completely satisfied, as the codec would (as you say) compensate for the added frame by compressing it down to near nothing (I had forgotten that it only encodes what has actually changed, guess it was a case of not seeing the forest for all the trees).

[fccHandler]

My DVD2AVI 1.76 doesn't have a way to select a framerate; there is only the option to "Force FILM" or not. (IIRC if you force film on hybrid material it looks very bad.) When you don't force film, DVD2AVI's output is always 29.97 (or 25 PAL) because DVDs don't allow any other rates. The NTSC portions are 29.97 interlaced and the FILM portions are 23.976 telecined to 29.97, due to the rff/tff flags in the stream.

[cheap-red-wine]

If you do not force film, you do not get duplicate frames in the film portions; you get duplicated fields, resulting in 2 interlaced frames out of every 5. If you use TomsMoComp on that you hurt the 3 progressives a little bit unnecessarily, but you totally ruin the two interlaced ones. If you want to retain all the good progressive frames and end up with a duplicate frame per cycle, you have to use a field matcher, such as Telecide(), Pfr(), or Uncomb().

[silver_cpu]

First off, I still don't understand what you mean by "destroying" the interlaced frames in the telecined material. I've found it an excellent filter for telecined material (except of course that it blurs things a bit), such as the new movie May which for some reason was telecined even though it was in theatres (I guess it's supposed to look better on TVs).

And, since I seem to be going about this the wrong way, would one of you suggest an avisynth file appropriate to my needs? I'm afraid after checking the FAQs on the telecine process I'm a bit lost as to what to do with the filters themselves.

[fccHandler]

Originally posted by silver_cpu
such as the new movie May which for some reason was telecined even though it was in theatres (I guess it's supposed to look better on TVs).
DVDs are made to be viewed on television, which is 29.97 fps. ALL film content is telecined (one way or another) when it is put onto an NTSC DVD.

would one of you suggest an avisynth file appropriate to my needs?
I use this a lot:
Mpeg2Source("mymovie.d2v")
Telecide(post=true)
Decimate(cycle=5)
It's not always the best way, but it's the easiest. ;)

[cheap-red-wine]

Originally posted by silver_cpu
First off, I still don't understand what you mean by "destroying" the interlaced frames in the telecined material. I've found it an excellent filter for telecined material (except of course that it blurs things a bit)Blurring a frame when you don't have to qualifies as being destroyed, doesn't it? :)

Here, I'll spell it out for you. Here is the situation before processing:

a a b c d
a b c c d

If you do field matching, you get:

a b c c d
a b c c d

There are 5 nice PROGRESSIVE, FULL RESOLUTION, NON-BLURRED frames. Of course, one is a duplicate.

If you deinterlace with blending the second and third frames will be blends. If you deinterlace with interpolation, you will lose resolution on the second and third frames. The amount of the loss will depend on the deinterlacer. With most deinterlacers (except FieldDeinterlace with an appropriate threshold), you'll also degrade the progressive frames.

Read the instructions for Telecide, Pfr, or Uncomb. They are all straightforward.

[silver_cpu]

Poop on quality. I'm using tom's mocomp, because it's easy, fast, and i really don't notice that much difference (i tried the other way and quite frankly it isn't worth my time to get such a small difference). I do want to sincerely thank both of you, however, for responding to my posts and patiently guiding me through the process. It's people like you who make this community run. I hope this thread helps others with the same problem.

[alucard83]

You can also read this (http://www.doom9.org/ivtc-tut.htm) ;)

[silver_cpu]

Egad...I've been through that FAQ and half a dozen like it in the past week or two. I finally sat down and did some serious visual testing, and to be honest there is only a *very* slight difference between the complicated, "properly" processed material and that which was filtered through Tom's Mocomp. Honestly, this really isn't worth the trouble. If I find a DVD where it is, then maybe I'll change my tune. 'Till then.