[XviD-devel] hq ac prediction mode

[suxen_drol]

On Sat, 8 Feb 2003 18:54:16 +0100 (CET) Christoph Lampert <chl at math.uni-bonn.de> wrote:

> On Sun, 9 Feb 2003, suxen_drol wrote:
> 
> > 
> > hi,
> > 
> > i implemented a new ac prediction calculation method, which considers
> > actual coefficient cost (bits), as ooposed to the present coeff sum
> > method. it reduces file sizes a little, though nothing significant.
> > 
> > 640x256, 500 consecutive i-frames
> > 	|		b	y	t	e	s
> > quant	|	coeff		bitstrema	saving
> > 	|	sum		sum
> > --------+------------------------------------------------------
> > q=2	|	8,105,984	7,938,048	167,936	(2.07%)
> > q=3	|	5,941,248	5,814,272	126,976	(2.14%)
> > q=4	|	4,812,800	4,718,592	94,208	(1.96%)
> > q=8	|	2,865,152	2,813,952	51,200	(1.79%)
> > 
> > 
> > 640x256, 500 frames (~10 i-frames, 490 p-frames)
> > 	|		bytes
> > quant	|	coeff		bitstrema
> > 	|	sum		sum
> > --------+---------------------------------
> > q=2	|	2,187,264	2,183,168
> > q=3	|	1,519,616	1,515,520
> > q=4	|	1,105,920	1,103,872
> > q=8	|	561,152		559,104
> > 
> > should i clean it up with a XVID_HQACPRED flag, and commit?
> 

commited. to enable hq acpred use:
	xframe.general |= XVID_HQACPRED;

> Sure... how was speed? 

for short sequences (500-1000frames) its unnoticble.
obviously that will depend on how many iframes are present in the
sequence.

i will perform some more extensive testing tonight.

-- pete; life is like a box of ammo