[MPlayer-users] Re: [-] Question on converting to DivX4...

[Rich Felker]

On Mon, Jul 11, 2005 at 02:40:11PM +0200, Matthias Wieser wrote:
> Am Samstag, 9. Juli 2005 05:03 schrieb RC:
> > On Fri, 8 Jul 2005 21:37:23 -0500
> >
> > Jeff Clagg <snacky at ikaruga.co.uk> wrote:
> > > You lose nothing! It just distributes it a bit differently.
> >
> > Either way you want to look at it (losing or redistributing), tfields
> > and pp=lb are about on even ground.
> 
> With pp=lb you lose halve the vertical resolution. With tfields=4 you lose 
> nearly no vertical resolution but you get smoother motion without 
> motion-blur effect.

Yes.

> pp=lb results in motion blur comparable to taking photos with an exposure 
> time of 1/50s or more. That's very bad.

Actually comparable to 1/30 sec, but it's even worse still.. rather
than blurring over the whole 1/30 sec interval, it only blurs from 2
points in that interval, each perhaps 1/120 sec in length. Thus you
have a ghost of two independent pictures, rather than a smooth blur of
a contiguous range of time.

> What about "kerndeint"? 
> Kerndeint only deinterlaces when it is necessary.

It's still a misguided strategy. You need to deinterlace to full
framerate or it will look like crap. Now, it's possible to deinterlace
with a better strategy than tfields does, trying to produce
full-height frames rather than just fields. This is a very heuristic
process and can go horribly wrong since the information actually
needed is _LOST_ and unrecoverable, but in practice it may be possible
to get good results reconstructing it through heuristics.

Rich