[XviD-devel] max_bframes>1 (was: Re: Quality optimization)

[Christoph Lampert]

On Fri, 24 Jan 2003, Felix von Leitner wrote:

> Thus spake Christoph Lampert (chl@math.uni-bonn.de):
> > Use a low max_bframes number, like 2 or 3. 
> 
> The ffmpeg quality benchmarks clearly showed that setting max_bframes to
> more than one was actually bad for quality.  So far, my xvid tests
> agree.  I have found substantial visual quality loss when setting
> max_bframes to 2 instead of 1.  I haven't tried 3 but I can't imagine it
> will be any better than 2. 

a) What I spoke about was a "brute force" algorithm to detect whether to
   use B-frames for a scene or not. The "max_bframes" would have been the
   _maximum_ possible number, not necessarily the number used in the end. 

b) ffmpeg tests were done with a rather low quantizer for bframes, 
   bquant_factor=1.25, bquant_offset=0.6 
   So for a fixed quant 4, the bframes would have quant 5, maybe 6. 
   My own tests with ffmpeg showed that a higher quant for bframes
   can saves enough bit such that the overall quality gets higher 
   (because reference frames have better quality). 
   I tested VQEG-PAL sequences with 0-3 bframes and different settings 
   for factor and offset. A max_b value of 2 won most often.
   And this was without dynamical setting! 

gruel