[MPlayer-dev-eng] mp-G2 pre18 / Video with no sound

[D Richard Felker III]

On Tue, May 06, 2003 at 12:37:47AM +0200, Arpi wrote:
> Hi,
> 
> > Also, altering duration is much easier than playing with timestamps, and for
> > containers having duration or fixed fps, it's preferred by timing code over
> > pts. Look at ported vf_tfields for sample.
> 
> Actually for your case (3:2 telecine removal) it's very easy:
> dmpi->duration=mpi->duration*4/5;

No, we wish it were that easy... In practice, telecine pattern is
often not regular. Causes may include scene change/splicing,
frame duplications to slow down video slightly post-editing, or mixed
telecine/progressive/interlaced content. When I set out writing this
stuff, I naively thought it would be simple, but it's a huge mess.

Anyway, am I correct in assuming that if I multiply all durations by
5/4, but don't drop *exactly* one frame out of every 5, A/V sync will
break?

If so, I think the simplest fix is to multiply frame durations by 3/2
when the frame has one of its fields repeated. This will correspond
exactly to the timing when watching the telecined video on a TV set,
but it won't be quite as smooth as the perfect 24 fps output would be.
So, I'd rather think of a nicer way to handle it.

BTW, about detc vs ivtc.... Yes the names are confusing, and yes they
both exist. They're two different filters that take different
approaches to removing telecine. vf_ivtc is a much nicer design, IMO,
and has optimized MMX code. The algorithm in vf_detc, however, may
work better when the telecine pattern is very regular but the input
signal is noisy (think analog TV input). Maybe for G2 I'll merge them
into a single filter with options to use several different algorithms.

Rich