[ExtractStream] Inverse 3:2 pulldown (was: Suggestions for Windows video editor?

[Scott Alfter]

On Thu, Jul 11, 2002 at 08:24:02PM -0700, Roger Merchberger wrote:
> --- Scott Alfter <yahoo@s...> wrote:
> > I usually create an Avisynth script to do A/V mux and NLE and run the
> > output of this script through VirtualDub for inverse 3:2 pulldown and
> > any filtering that might need to be applied. You can then reencode the
> > resulting video with TMPGEnc.
> 
> You'd mentioned the inverse 3:2 pulldown before -- it's to create a
> 24fps (or 23.997 or whatever that is...) file, right?

23.976 fps (29.97*4/5).

> I've not used your system, but I tried that inverse 3:2 pulldown in
> TMPGenc, 2470 kbps bitrate, and encoded... The resulting .mpg file, while
> it seemed to play well enough on my machine (didn't test it on an SVCD...)
> wasn't any smaller. So... I have a few "3:2 newbie" questions:
> 
> 1) What advantages do you see with using the inverse 3:2 pulldown?

The file won't be any smaller, but more bits will be allocated to each
frame. Consider a 2-Mbps video stream, one encoded at 30 fps and the other
at 24 fps (the numbers are rounded to simplify the math). To simplify the
comparison, we'll say we're only doing I-frame encoding (a la MJPEG). 2e6
bits divided across 30 frames means you get about 8.1 kB to encode each
frame. The same number of bits divided across only 24 frames gives you
about 10.1 kB for each frame. Each frame can thus be encoded at slightly
higher quality.

(In real MPEG-2 encoding, of course, you have P and B frames to deal with. 
These will be smaller than I frames, but given a constant bitrate, a lower
framerate should allow more information to be stored in each frame.) 

If it's properly done, you also eliminate nearly all interlacing from the
video. While MPEG-2 does support interlaced video, progressive-scan video
is easier to encode and will look nicer when played back on a
progressive-scan display (such as your computer monitor). Encoding
four-fifths as many frames will also help speed up encoding.

> 2) Logic would say: using the same bitrate, a slower fps file should
> be smaller, right?

At the same bitrate, the file size for a given length of video should be
roughly equal. The difference is in how the bits are allocated to each
frame. Dividing a gallon of ice cream among 24 people means everybody gets
more ice cream than if you divide the same gallon among 30 people.

> 2a) Does TMPGEnc "do the right thing" or does 3:2 pulldown not
> work with that software?

Several programs (TMPGEnc, VirtualDub, Avisynth, etc.) implement inverse 3:2
pulldown. Of the ones I've used, the adaptive inverse 3:2 pulldown in
VirtualDub is the easiest to use and produces the best quality.

> 3) Do you know if most DVD players that play SVCDs handle 24fps SVCDs
> or is it a "try and see?"

Given that many (most?) DVDs are encoded at 23.976 fps (rip one and split
the VOBs with DVD2AVI to see for yourself), I'd expect that most of the ones
that play SVCDs should handle SVCDs at that framerate. I think that support
for 23.976 fps is part of the spec for MPEG-2 and/or SVCD, but don't quote
me on it. :-)

_/_
/ v \ Scott Alfter (salfter@s...)
(IIGS( http://salfter.dyndns.org
\_^_/ rm -rf /bin/laden