lorenm at u.washington.edu
Sun Apr 4 02:07:40 CEST 2004
On Fri, 2 Apr 2004, D Richard Felker III wrote:
> On Fri, Apr 02, 2004 at 10:25:50AM -0800, Loren Merritt wrote:
> > The reason (I infer) that XviD can fill any desired bitrate, and lavc
> > can't, is that XviD supports quantizer 1 (-xvidencopts
> > min_iquant=1:min_pquant=1). lavc has a vqscale=1 (constant quantizer), but
> > doesn't respect vqmin=1 in 2-pass encodes.
> You need lmin too. Then it should work fine.
> > Despite what the lavc entry in the manpage says about quant 1 making
> > little quality difference, XviD with min quant 1 looks much better than
> > lavc when lavc maxes out with quant 2 at 800kbit/s (and I am encoding at
> > unscaled dvd resolution).
> If you're using 800kbit/sec, quant 1 is useless. It will just waste
> more bits on frames that already look perfect and then your other
> frames will get higher quantizers.
> In fact, 800kbit/sec is so low for full DVD resolution, that I'd be
> surprised if any quantizers come out less than 4 or 5, and most are
> probably 7-9 or higher!
> The key to good quality is proper cropping and scaling, not using full
> DVD res...
That 800kbit/s comes from encodes using fixed quant 2, mplayer 1.0pre3:
mencoder -o neko.avi dvd://2 -mc 0 -noskip -skiplimit 0 -nosound
-vf crop=672:432:24:22,hqdn3d=1 -ovc lavc -lavcopts vcodec=mpeg4:
or similar quality and bitrate from xvid 1.0pre3:
The movies involved are "Haibane Renmei" and "Neko no Ongaeshi", which are
animated at around 10fps (variable framerate), and contain not all that
much action. At quant 2, Haibane is 820 kbit/s, and Neko is 910 kbit/s.
Still, frames at quant 2 are not perfect, and they look much better with
(And now that you mention it, vqmin=1:lmin=1 also works.)
More information about the MPlayer-users