[MPlayer-users] low bitrate/high quantizers can give good result with lavc (?)

[D Richard Felker III]

On Wed, Sep 24, 2003 at 10:46:14AM +0200, HR wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> 
> I just encoded a movie, and after the first pass, my script reported
> 
> 09/23/2003 22:13:41  mencoder movie.vob -chapter 1- -oac copy -ovc lavc
> -lavcopts
> vcodec=mpeg4:keyint=125:vbitrate=949:mbd=2:v4mv:precmp=2:cmp=2:subcmp=2:
> vpass=1 -sws 2 -vop scale=640:272,crop=712:416:6:80 -o /dev/null
> 09/24/2003 02:29:16  1st pass complete, listing quantizer distribution
> 09/24/2003 02:29:18  q1 3.7%
> 09/24/2003 02:29:18  q2 5.5%
> 09/24/2003 02:29:18  q3 24.5%
> 09/24/2003 02:29:19  q4 23.2%
> 09/24/2003 02:29:19  q5 18.0%
> 09/24/2003 02:29:19  q6 10.9%
> 09/24/2003 02:29:19  q7 6.3%
> 09/24/2003 02:29:20  q8 4.5%
> 09/24/2003 02:29:20  q9 3.1%
> 09/24/2003 02:29:20  q10 1.5%
> 09/24/2003 02:29:20  q11 0.7%
> 09/24/2003 02:29:21  q12 0.4%
> 09/24/2003 02:29:21  q13 0.2%
> 09/24/2003 02:29:21  q14 0.2%
> 09/24/2003 02:29:21  q15 0.2%
> 
> To my slight surprise, the result looks pretty good, even at this

Of course a bitrate of 949 should look good. Normal bitrate for a 1cd
encode is around 650-750, which also looks very good for most
material.

> relatively low bitrate and the reported quantizer distribution. A little

At 640x???, quantizers up to 4 or 5 don't look bad at all, and even up
through 6 or 7 they won't look bad except during low motion/stills.

> bit of ringing, and I have to scale it up to make it significantly
> annoying. I think the only option to improve this result is to increase
> the bitrate, since I'm already downscaled to 640:272. Right?

hqdn3d! Watch it magically improve quality by an order of magnitude
with no effort! :)

Also your keyint is too low (250 is default and 300 should be ok) and
you didn't enable trellis quantization.

Finally, it seems you're looking at the results of pass1, which are
not very helpful in determining final quality. You should look at
results of pass 2 instead!!

Rich