[MPlayer-dev-eng] Re: ccache

[Michael Niedermayer]

Hi

On Monday 06 January 2003 03:52, Arpi wrote:
> Hi,
>
> > no, iam not joking, this is exactly what the mpeg4 & mpeg2 std says for
> > interlaced YCbCr 420
>
> btw, how are these handled in mplayer? i guess they aren't.
yes, we ignore it, allthough for some cases it would be relatively easy to 
handle correctly if we would know which variant was used

> at least i can't imagine how does postproc or swscale or even teh vo
> drivers know if that YV12 frame comes from jpg/mpeg1, mpeg2 (frame or
> fields) or mpeg4? since they have different chroma rules...
hmm, perhaps we should do 2d cross correlation (with subpixel accuracy) of the 
luma & chroma planes ;)

>
> also, what about the 4x4 subsampled YUV 4:1:0 (YVU9) used in indeo &
> sorenson? both rules (c't's and the pdf's) description says that 4:1:0
> means 4x2 pixels subsampling (4 Y 1 UV in even lines, and no UV in the odd
> lines) but it's really 4x4 (UV only in every 4th lines). and where are UV
> pixels sampled there?  
for sorenson i think its in the middle of the 4x4 luma and for indeo its at 
the first sample, but thats purely from my memory of how they looked when i 
implemented YVU9 support

> at least i can differenec displaying lwores YVU9
> using ati mach64's hw YVU9 scaler and swscaler (swscaler looks much better
> :)), ati's one shows chroma a bit shifted.
>
> and anyway, what does the whole chroma subsamplin means ?
> there are 4 pixels:
> P1 P2
> P3 P4
>
> is U the average of P1..P4's U values, is the average of P1 & P3's average,
> is P1's U or what?
i guess its upto the encoder/ capture hw / whatever to decide but ideally the 
chroma plane should simply be scaled & shifted with a good filter (cubic or 
whatever)

[...]

Michael