[MPlayer-users] lavc-Options for *BEST* quality?

[D Richard Felker III]

On Sun, Feb 02, 2003 at 04:08:07PM +0100, Sven Over wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> Hi, all.
> 
> I"m affraid I'm not up-to-date with all that lavc-options. (E.g. what is 
> qpel?)
> 
> So, my question to the experts in this list: what are the options to use with 
> lavc if I want to get highest possible quality for a given bitrate. CPU-time 
> (almost) does not matter.
> 
> Thanks in advance.

Rémi (on the ffmpeg list) has posted some very interesting results
from extensive tests of all the nice encoding options, which you can
read about at http://rguyom.ath.cx/. Keep in mind these are optimized
for maximizing PSNR (an objective quality measurement, although not
necessarily the best for making the movie "look good").

So far, the winning combinations all seem to include vhq:v4mv:trell.
Adjusting *cmp (subcmp, cmp, precmp) can also make a big difference.
For anime, Rémi found 1 to be best, while for real-world content, 3
was the best with 2 being almost as good and much faster to encode.

Also, some discussions of the matter in the ffmpeg lists suggest that
using 2 or 3 for the subcmp while using a simpler comparison (0 or 1?
I forget...) for cmp and precmp might actually produce better results
in some cases.

And finally, if you're trying to get nice quality with low bitrate
(e.g. a 1-cd dvdrip of a movie > 80 minutes long), you should also try
the adaptive quantization options (naq, *_mask) and single coefficient
elimination (vcelim, vlelim). These actually reduce quality in parts
of the image to save bits for where they might be needed more.

Rich