Re: YUV format

Holger Waechtler <[email protected]> Thu, 26 Dec 2002 23:25:03 +0100
Newsgroups gmane.comp.multimedia.ogg.tarkin.devel
Message-ID <[email protected]>
Ed Okerson wrote:
> On Thu, 26 Dec 2002, Holger Waechtler wrote:
> 
> 
>>After all you should remember that this code is for research, it
>>compresses oonly in statial domain and does not exploits any temporal
>>coherence.
>>
>>So you have to compare it to still image comression algorithms, not to
>>video codecs unless you add a motion compensation scheme before.
> 
> 
> I thought 3D wavelets included the temporal domain as the 3rd dimension.

You can enable this by defining N_FRAMES in tarkin.c to anything > 1. 
IIRC it was hardcoded to 1 because the time domain transform did not 
gained much - we saw that a 'real' motion compensation would be needed.


> I switch to a different source file, and it is more obvious what is
> happening.  It appears first as some random noise on the picture and then
> the noise level goes up exponentially until it is nothing but random noise
> over the entire image.

Hmm, I just was running a few frames from the venuscubes test sequence 
again, I had about the same results as I remembered:

First: you want to modify the CFLAGS in the Makefile to enable 
optimisations (-O3 instead of -O0) and disable debug checks and 
assertions. Then I was running something like this:

$  ./tarkin_enc <path>/AnimSpace0000%01d.ppm 3000 4 4

which means that we encode the first 10 frames from venuscube sequence 
at about 3kBytes per encoded frame using (4,4)-wavelets. When I piped 
the status messages to /dev/null it needed about 2 seconds on my old 
vaio at ~500MHz.

I don't remember anymore if the latest coefficient encoder was designed 
for speed or compression. Compared to a JPEG at the same bitrate the 
pictures look a bit smoothed but ok. I did not measured PSNR, you can do 
this on your own...

Decoding:

$ time ./tarkin_dec stream.ogg &> /dev/null

yields about 1.1 seconds decoding time. The venuscube sequence and some 
other test sequences are available at http://media.xiph.org/.

Holger

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to '[email protected]'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.