[MPlayer-users] Encoding problems / tips needed

Loren Merritt lorenm at u.washington.edu
Tue Aug 3 22:12:49 CEST 2004


On Mon, 2 Aug 2004, JiHO wrote:
> Loren Merritt wrote:
>> On Sun, 1 Aug 2004, JiHO wrote:
>>> even using nearly all the "quality providing options" of lavc
>>> (if you think of others, please do not hesitate)
>>
>> vmax_b_frames=1 (usually significant quality gain, but has some
>> side-effects, including increased cpu time to decode (+5% on my computer),
>> and reduced compatibility with some hardware players)
>>
>> preme=2:predia=-4 (helps almost as much as 'dia', and is much faster)
>
> thanks again for this. one problem though: the motion estimation prepass
> related to these options should be done at first pass? second pass? both?
> thank you in advance.

All options (except vpass itself) should be identical on both passes.
(It's also ok if bitrate is slightly different: vqscale on the first pass
sometimes gives me better results, and has never been worse than vbitrate
as long as I pick a qscale that ends up somewhere near the 2nd pass
bitrate.)

Maybe the name "ME prepass" is misleading. It has nothing to do with 2
pass encoding. It means to run motion estimation twice per frame. In this
case, that's first with a large search diamond (adaptive size 4) and fast
comparison function (SAD), and then with a smaller diamond (normal size 1)
and better cmp (DCT).
I have found those particular values to be a good tradeoff between
quality and speed, but it depends on the video content. YMMV.

--Loren Merritt




More information about the MPlayer-users mailing list