[MPlayer-users] Q: libavcodec, B-frames, mbd, v4mv

[D Richard Felker III]

On Mon, Aug 25, 2003 at 05:22:30PM +0200, Michal Seliga wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> > Date: Mon, 25 Aug 2003 16:16:54 +0300 (EEST)
> > From: Tuukka Toivonen <tuukkat at ee.oulu.fi>
> > To: mslama at email.cz
> > Cc: mplayer-users at mplayerhq.hu
> > Subject: Re: [MPlayer-users] Q: libavcodec, B-frames, mbd, v4mv
> > Message-ID: <Pine.GSO.4.55.0308251610110.10178 at stekt37>
> > In-Reply-To: <3F4A06E2.000001.30597 at tajfun.atc.cz>
> > References: <3F4A06E2.000001.30597 at tajfun.atc.cz>
> > Content-Type: TEXT/PLAIN; charset=US-ASCII
> > MIME-Version: 1.0
> > Precedence: list
> > Reply-To: "MPlayer user's list." <mplayer-users at mplayerhq.hu>
> > Message: 4
> 
> > On Mon, 25 Aug 2003 mslama at email.cz wrote:
> 
> > >besides CVS log? B-frames are disabled by default. Any notes/suggestion
> > >about enabling them? (ie. what is probable gain of using them?)
> 
> > Very little. I was recommended to use a single B-frame at maximum. Anyway
> > the difference is so small that I wonder whether it's a libavcodec-specific
> > problem or why the standards include B-frames anyway?
> 
> I disagree. In some cases b frames help a lot. mostly in lowmotion scenes. but they

In theory they help the most during fades or when you have fast moving
objects in front of a mostly-still background. However, no encoder is
actually able to make intelligent decisions for how to use B-frames,
so they're not very useful in practice...

> have one big disatvantage - classic divx codec on windows will not play such files
> correctly (you would have to use ffdshow codec, which is IMHO better anyway). and
> sometimes (very noisy or low quality encoded source) it can make video bigger (with 
> same quality) or uglier (with same size)

This is more common than you would think...

> But I agree that single B-frame is usually enough... (more b-frames can make file shorter
> but difference is not so big compared to difference between 0 and 1...)

Remi's tests showed that more than one B-frame decreases quality, so I
would definitely not recommend more than one! Also in his tests, even
using 1 B-frame reduced quality with anime (but not with live action
movies), so I'm skeptical of B-frames alltogether.

> What I usually do is that I encode video using all possibilities for b-frames (0-4) with fixed
> quantizer and for final encoding I use settings which made the output file shortest.

This is not a good way of comparing, since constant quantizer will
waste tons of space on scenes which are essentially stills.

> (don't laugh, I prefer quality before encoding time ;-)

Preferring quality is ok, but you should understand when the quality
really makes a difference, and when it doesn't...

Rich