[MEncoder-users] lavc or xvid

Corey Hickey bugfood-ml at fatooh.org
Sun Aug 26 20:22:03 CEST 2007


RC wrote:
> On Sat, 25 Aug 2007 15:46:48 +1000
> Alex Samad <alex at samad.com.au> wrote:
>> vcodec=mpeg4:threads=2:mbd=2:mv0:trell:v4mv:cbp:last_pred=3:predia=2:
>> dia=2:vmax_b_frames=2:vb_strategy=1:precmp=2:cmp=2:subcmp=2:preme=2:q
>> ns=2:aspect=16/9:dc=10:vbitrate=$VBITRATE
> 
> *cmp=2 makes for quite a bit of blockiness, and is usually a bad idea.

Under what circumstances? I have not observed this, though I admit I 
haven't done any testing of the *cmp options in a long time. A long time 
ago, *cmp=2 helped my tests' PSNR significantly and made the encodes 
look a little better.

> It'll look much better without *cmp, or perhaps with 3 instead of 2. 
> qns also only seems noticably to help with animation.

...and in encodes where the bitrate is somewhat low. qns reduces 
blockiness around sharp edges, and higher bitrates reduce blockiness 
everywhere, so the advantage of qns is negated once bitrate is 
sufficiently high.

> You're usually better off with 0 b-frames.

I disagree. In my experience, 2 B-frames is optimal. The only place 
B-frames should hurt quality is in high-motion scenes, and an 
appropriate vb_strategy will avoid that. On that subject, I should note:

* vb_strategy=2 gives the best quality, but it is very slow
* vb_strategy=1 works better with b_sensitivity=10 or so

I wrote stuff about this a while ago:
http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2006-May/011523.html

-Corey



More information about the MEncoder-users mailing list