[x264-devel] Re: h->stats

[Måns Rullgård]

Loren Merritt <lorenm at u.washington.edu> writes:

> On Fri, 29 Apr 2005, Jeff Clagg wrote:
>> On Fri, Apr 29, 2005 at 09:12:00AM -0700, Loren Merritt wrote:
>>> I've seen gains of around 4 dB from using a better 1pass algo.
>>
>> Is this a typo? It's hard to believe the old cbr could be so bad.

The old CBR is still infinitely better than constant QP, if you have
to be VBV compliant.

> No, it's not a typo. You won't see that much difference in the
> standard testclips, since they tend to be short and constant
> complexity. But when the content contains any sort of variation, CBR
> is very bad. My +4 dB figure is for (average, I hope) movies, with a
> target average bitrate but no strict VBV constraint.

I wrote the current CBR to stay within VBV constraints, simply because
that was the requirement.  The stream had to work with low latency,
and little headroom of link bandwidth.  If it can be done better, go
ahead and replace it.

-- 
Måns Rullgård
mru at inprovide.com

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html