[XviD-devel] better rate control

[Marco Al]

Dirk Knop wrote:

> My conclusion is that redistributing the rounding error while
> downscaling the bitrate curve is giving us bigger deviation from the
> target filesize (the rounding error was 89kb when using my statsreader)
> - but:

A leaky bucket approach will get you pretty close to your target
filesize no matter how wrong your per frame predictions are.

> I don't see where there is urge for "better fit to target filesize".....

It's more aimed at getting the per frame bitrate right, without too
much over/undershoot during encoding.

To me it seems a nice experiment with it would be to rip out
the curve compression based on bitrate entirely, and use a
R-D curve instead using rate and distortion predictions from
the p-domain method (I use the term R-D curve in a slightly
non conventional way, with it I mean a curve which gives the
allowable rate at a given distortion ... so for constant
distortion coding you would give a flat line for instance).

Marco