[XviD-devel] reduced resolution VOPs (rrv) and all sort of things

[Marc FD]

> the rrv stuff: basically, the error correction
> *only* is transmitted at half the resolution. That
> is after iDCT and 8x8->16x16 upsampled you've
> regained the full resolution. Motion prediction
> is not at half resolution, and this is a cool
> thing: you can still take advantage of the
> previous, detailed, frames. 

so it's texture scaling. i really want to see
how it performs. it think i could be usefull for 
(bitrate eating) fast scenes .

> B-VOP are not subject to
> rrv coding (I wonder why...). 

for me it makes sense...
for low motion, where you can notice a lot of 
details, bvops have very accurate prediction
and they'll never be used as references.
for high motion, reduced-ivops will contain
a _lot_ of texture data (many intraMBs and
complex interMBs) so scaling the texture
will reduce a lot the texture data to encode,
but it'll be somehow better than using too
big quantizers, because following frames
will have better prediction.

i'm totally wrong or it makes sense ?

> My feeling
> is that this would be useful for fast-moving
> scenes to alternate full/half P-VOP,

that's my feeling too. (full/quarter vop ^^)

> not
> to spoil the bitrate with transient details.
> I roughly tested the idea, but I have nothing
> conclusive as of now. Ok, I'll try to setup a
> binary of my toy codec next week, with rrv 
> stuff in it, so people can try to have an
> opinion by theirselves...

that would be cool, to wait until pete implement
it in XviD, and to see if it as promising as it seems.