Re[2]: [XviD-devel] rrv en/decoding

[Marc FD]

> > Surely all the reference frames are stored in there 'native' format, and
> > only the transform and motion vector decoding need be altered for
> > reduced resolution. See 7.6.10
>
> So the b-vops are in reduced resolution, too? If it's so, in what
> macroblock modes the vectors are in half-zero precision? (forward,
> backward and interpolate surely, but what about direct?)
>
> I see that video verification model doesn't mention it at all.

i think it's mainly because no profile allows both rrvs (that's in adv real
time simple) and bvops (in adv simple)

BTW, i just tested rrvs with current CVS. it's different from what i
expected it too look,
but it performed very well on my torture sample (10 sec of pure movement)
because rrvs are smaller at quant 5 than normal vops at quant 10. and they
look much better, and the extreme smoothness is not really a problem because
of the speed of motion.

pete, SysKin, good job  !!

i'm hacking a lot the encoder to test various ways of using this(too?)
powerfull tool,
i'll reports the results if i find good ways of using it.

Cheers,
MarcFD