[MPlayer-DOCS] CVS: main/DOCS/man/en mplayer.1,1.1186,1.1187

[Corey Hickey]

Guillaume Poirier wrote:
>>CVS change done by Guillaume Poirier CVS
>> .TP
>>+.B bidir_refine=<0\-4>
>>+Refine the two motion vectors used in bidirectional macroblocks,
>>+rather than re-using vectors from the forward and backward searches.
>>+This option has no effect without B-frames.
>>+.PD 0
>>+.RSs
>>+.IPs 0
>>+Disabled (default).
>>+.IPs 1\-4
>>+Use a wider search (larger values are slower).

Thanks.

> While I'm thinking about it, is it desirable not to activate this
> feature by default?

I haven't tested it enough to say how much it helps yet. I tried running
some short clips, but they were so short that the bitrate varied so much
to make the PSNR nearly meaningless. The impact on speed is odd -- it
slows down both passes, but the first pass more so.

Here's the encoding command, with fairly simple lavcopts:

for i in 1:turbo 2 ; do
  time mencoder ../serenity_m420p.mov -nosound -ovc lavc -lavcopts \
vcodec=mpeg4:psnr:mbd=2:mv0:trell:cbp:vmax_b_frames=2:bidir_refine=0:vpass=$i
  -o 0.avi ; done

Here are the user times obtained for each pass. Note that I use turbo
for pass 1, which makes the first pass faster at low values of bidir_refine.

bidir_refine=0
2m31.590s
4m3.825s

bidir_refine=1
3m34.203s
4m45.861s

bidir_refine=2
6m28.443s
6m21.001s

bidir_refine=3
10m26.342s
8m31.295s

bidir_refine=4
12m27.799s
9m38.598s

Also, for what it's worth, these times are from a P4; I suppose it's
vaguely possible an Athlon won't have such odd results.

> I'm trying to find what could be saner defaults for lavc MPEG4, and this
> option, depending on its impact on speed and quality could be one of
> those (like mbd=2 and trellis) which are to be on by default.

Once I finish testing Michael's other changes then I'll clear up some
disk space and test bidir_refine with The Matrix.

-Corey