[MEncoder-users] new doom9 codec comparission (submission)

[Michael Niedermayer]

Hi

On Sun, Dec 11, 2005 at 06:31:48PM -0800, Corey Hickey wrote:
> Michael Niedermayer wrote:
> >>>>Without being able to test your source directly, I would suggest using
> >>>>vqscale=1.5 instead of vqscale=2.15. Don't take that to be my final
> >>>>answer, though; I'm hoping to get some insight from other people on
> >>>>mencoder-users. vqscale=1.5 is based on looking at a graph I have of
> >>>>quantizer vs bitrate for snow, and erring on the low side (1.8 is my
> >>>>closest guess, but 1.5 should be safe).
> >>>
> >>>
> >>>Well, you can use AviSynth with Wine, and decode the AviSynth script via 
> >>>avs2yuv.. I think dgindex also works on Linux if you don't have a windows 
> >>>box to test.
> >>
> >>I would try to do that, but I really don't have time right now... :(
> >>I'll see if I can tomorrow, but it might be too late for you.
> 
> Well, I managed to get my laptop (running windows right now) set up
> encoding from the avisynth source, using the same script and parameters
> as Doom9. It's going to be a few more hours before it's done, but I
> don't think there will be a significant difference in bitrate from the
> test encodes I did under Linux. I've been "following" my laptop encode
> with my desktop machine, and at all the intervals I checked the bitrate
> was the same +/- a few percent.
> 
> Assuming that I've reproduced the same environment as Doom9, that likely
> means the DVD I used is different, somehow. Maybe it's more grainy.

one way to check that would be to cut out a short part of the mpeg2 bitstream
of both and compare them in some player side by side ...


> 
> > i would guess that lmin was too large, though its not clear how it could
> > have worked for corey ...
> 
> Does lmin have any effect on the constant-quantizer first pass? I'm

hmm, IMHO no


> figuring that if the first pass hadn't used such a low bitrate, then the
> second pass would have used the specified bitrate.
> 
> Doom9 pass 1: 479.429
> Doom9 pass 2: 519.393
> 
>    my pass 1: 592.388
>    my pass 2: 581.367
> 
> 
> Right now, I'm testing whether it would work fine to ask Doom9 just to
> use vqscale=1 for the first pass, which is way lower than what it needs
> to be. If that ends up working fine, it's an easy solution that also has
> the benefit of not needing to be adjusted for different encodes (except
> for very-high-bitrate ones).
> 
> Any comments?

if qscale=2.15 is significantly below the target then =2 will be too, and
lmin has a default of 2 IIRC so i dont think the second pass will reach
the target size no matter what you change in the first pass

[...]
-- 
Michael