[XviD-devel] qpel bugs (was: xvid decoding fails)

suxen_drol xvid-devel@xvid.org
Sun, 15 Dec 2002 15:35:14 +1100


On Fri, 13 Dec 2002 09:01:00 +0100 "Frodo" <csharprules@hotmail.com> wrote:

> Hi guys,
> 
> I've downloaded the latest CVS tarball and noticed that now my older xvid
> movies
> got big problems. It's hard 2 explain, but it looks like the following:
> - the Y plane is ok
> - the U/V plane seems to drift around the screen a bit, but only on areas
> which have high
>   U/V itensity
> 
> for example I got a scene which shows a small red light standing on a desk
> the light itselfs stays steady, but the red light it produces moves around
> it

your video was compressed with quartpel enabled. quarterpel is still
experimental, and we're trying to sort out some bugs present either in
xvids encoder, or in other decoder.

using todays development libxvidcore, i didnt "see" any chroma errors
whilst decoding your clip. anyone else care to comment?

> When I'm using an older xvid decoder (like the one from the ffmpeg group)
> the scene is played fine

i just ran some qpel interop tests (using a chunk of natural video), and
it seems:
* divx5.0.2 displays nasty Y and UV artifcats whilst decoding xvid+qpel
content
* envivio1.2.46 "appears" to decode xvid+qpel content correctly
* envivio1.2.46 "appears" decode divx5.0.2+qpel content with UV
artifacts.
( i say appears, because iam using graphedt to use envivio)

using radeks artifical qpel test.avi, it seems:
* xvid decodes it fine
* divx5 display HV interpolation errors
* envivio1.2.46 displays H, V,and HV interpolation errors (argh!)

wtf do we do about this?

-- pete; life is like a box of ammo