[XviD-devel] VHQ

[Christoph Lampert]

Hi,
me again, for the last time, promised: 

I forgot to do the most important check: just mode decision,
no DCT-refinement (inspired by vhq mode in ffmpeg): 
foreman, maxb=2, bquant_ratio=2.00, etc. 

BITS-mode-decision, Fixed Quant 4
Q0 s 1779 e: 79.3 , d: 334.5 P(240): 36.27 B(154): 36.24 
Q1 s 1306 e: 70.9 , d: 328.5 P(215): 36.72 B(179): 36.51 
Q2 s 1306 e: 63.3 , d: 291.9 P(215): 36.72 B(179): 36.51 
Q3 s 1217 e: 56.5 , d: 315.4 P(220): 36.77 B(174): 36.53 
Q4 s 1175 e: 48.6 , d: 315.6 P(217): 36.80 B(177): 36.48 
Q5 s 1012 e: 31.5 , d: 147.5 P(220): 36.94 B(174): 36.84 
Q6 s 1004 e: 29.3 , d: 147.4 P(223): 36.97 B(171): 36.82 

BITS-mode-decision, bitrate 128 
Q0 s 630 e: 129.2 , d: 510.1  P(384): 29.52 B( 10): 32.19 
Q1 s 477 e:  87.9 , d: 421.5  P(284): 32.00 B(110): 32.52
Q2 s 477 e:  87.5 , d: 427.6  P(284): 32.00 B(110): 32.52
Q3 s 474 e:  70.8 , d: 416.0  P(282): 32.28 B(112): 32.81
Q4 s 469 e:  56.6 , d: 416.6  P(279): 32.40 B(115): 32.82
Q5 s 446 e:  37.0 , d: 175.3  P(266): 32.94 B(128): 33.34
Q6 s 440 e:  32.8 , d: 173.0  P(262): 33.03 B(132): 33.43


If you compare these two to the previous mail, you see the real thing: 
For fixed quant, PSNR gets slightly _higher_ while bitrate drops (not as
much as for BITS-refinement, but significant). 
For bitrate mode, PSNR is clearly higher than no-BITS or BITS-refine,
while speed drops much less (not by 50% less fps, but only 10-20%)

gruel