[MPlayer-dev-eng] Re: [patch] lavc ME parameter handling cleanup & 2 little fixes

[Rémi Guyomarch]

On Sun, Apr 21, 2002 at 10:34:13PM -0400, D Richard Felker III wrote:
> On Mon, Apr 22, 2002 at 03:09:10AM +0200, Rémi Guyomarch wrote:
> > Hi,
> > 
> > I'm cross-posting since this patch touch both mplayer and
> > libavcodec. Hope you don't mind ...
> > 
> > For libavcodec :
> > - move Motion_Est_QTab[] to mpegvideo.c, since there's no need to
> >   store it as many times as one include "avcodec.h" ;
> > - use symbolic names instead of numbers in this array.
> > 
> > For mplayer :
> > - use Motion_Est_QTab[] from libavcodec so that the "vme" parameter is
> >   now quality-oriented (0 = no ME, 1..5 = fastest to slowest, worst to
> >   best). IMHO it's way more intuitive than the previous ordering ;
> > - don't use 4MV with B frames as it's buggy right now.
> > - don't use 4MV in "-pass 1" since it's not supported by lavc
> >   (same problem as HQ mode). Symptoms : libavcodec refuse to allocate
> >   context in MPV_encode_init() and mencoder crash at the fist call to
> >   MPV_encode_picture().
> > 
> > PS: You are free to tell me that I'm on crack :-)
> 
> You are on crack.
> 
> Seriously, don't reorder the ME methods. Your reordering doesn't even
> make sense. AFAIK, LOG and PHODS are considerably slower AND lower
> quality than EPZS! FULL may or may not be better than EPZS.

I took this order directly from avcodec.h, I didn't invented it.

And yes, I didn't tried every ME algo, but IMHO if one is slower *and*
worst than EPZS, it should be simply deleted. Why keep it ?

-- 
Rémi