[MPlayer-dev-eng] DivX5 & VBR encoding

[Kim Minh Kaplan]

D Richard Felker III writes:

> On Mon, Sep 09, 2002 at 03:08:50AM +0200, Kim Minh Kaplan wrote:
>
>> Yes, pass one does constant quantizer (2) encoding, so you better
>> direct the file to /dev/null.
>
> Is there a good reason for this? I'm not terribly familiar with 2pass
> encoding theory aside from the basics, but I know the way lavc does it
> is quite different. Are there papers on the issue?

IBM Journal of Research and Development published "Two-pass MPEG-2
variable-bit-rate encoding".  See
<URL:http://researchweb.watson.ibm.com/journal/rd/434/westerink.html>.
That's the only interesting one I found so far.

When you read this paper, the quantizer of the first pass has no real
importance except the it is used to calculate the complexity of each
frame with complexity=bits*quantizer.  An argument could be made
stating that when you let the encoder internale rate controler choose
the quantizer (as in divx4_vbr) you essentially state that you have no
real control about what you'll get whereas when you say quantizer=2
you have at least something that is constant across all movie.

I have read and understood how divx4_vbr works (it's mostly what is
described in IBM's paper).  I once tried to read xvid_vbr but it is
much more complex and have not yet understood it.  Now how does lavc's
VBR work?  Would it be usable for DivX?

Kim Minh.