[MPlayer-dev-eng] [RFC] change lacv's MPEG-4 encoding options defaults

Rich Felker dalias at aerifal.cx
Wed Dec 28 18:03:26 CET 2005


On Wed, Dec 28, 2005 at 02:48:45PM +0100, Guillaume Poirier wrote:
> Hi there,
> Wouldn't it be nice if lavc's default's were the best speed/quality
> tradeoff?
> 
> Currently, lavc's defaults are pretty much "blazing speed at all costs",

They're not "blazing speed at all costs". They're "no options
enabled". This is the standard for MPlater.

> which I don't think is a good idea since it doesn't help promoting lavc
> as a high quality codec, but just "good very fast codec".

IMO it's worse for promoting lavc if we provide defaults which are
supposed to be good and they perform badly for someone.

> I'd like lavc to default to vcodec=mpeg4:mbd=2:trell:v4mv:turbo

turbo?!?! WTF!! This is absolutely not acceptable. Many users, myself
included, use vpass=1 when doing 1-pass encoding just to read the
stats file or optionally be able to do a second pass if it looks bad,
when making files with constrained bitrate.

Also, v4mv?? Afaik there's no way to turn it off, and it's definitely
not always wanted, e.g. for snow.

> I'd like to consider adding vmax_b_frames=1, vb_strategy=2 and dia=-1,
> but I have not made any recent speed/quality tests recently.

B frames? Never! There are too many cases where they CANNOT be used,
like muxer_lavf, and currently they cause a/v desync if you use
MEncoder's a/v sync code.

> Maybe with latest Michael's commits, further tests should be done to
> find the new best speed/quality tradeoff, but until then, what do you
> guys think?

This topic has come up in the past and I've flamed in the past.
Defaults should always be all-off. If users are too lame to RTFM they
should not hope to get a good encode.

Rich




More information about the MPlayer-dev-eng mailing list