[theora] A comparison of VP3, and two MPEG-4 variants

[Christoph Lampert]

On Tue, 25 Mar 2003, Colin Mckellar wrote:
> 
> Interesting. I will look at those. However, since they are interlaced, 
> that is a serious limitation. 

Simply drop every second line (the clips are in YUV 4:2:2, so
there's no chroma problem), and if it matters, also drop every second
pixel to get aspect ratio right. 
This can easily be done on any plattform, so the results will be
comparable, whereas most de-interlacing methods would not. 
I did this and tested with XVID... the clips are still challenging, even
with reduced res. 

With an automatic script for different sequences and different encoding
options, I got some nice tables (outdated by now), like 
http://florin.myip.org:8000/users/chl/VQEG-27-01-03-branch.html

> None of the codecs that I used support 
> interlaced video.. so testing them with interlaced video is unrealistic.
> I believe that interlacing in this day and age is unnecessary. 
> Interlacing really only has a place on old school TVs.

And by far too many modern DV cams :-( 
 
> > Without preprocessing, scaling, cropping those ckips should give
> > PSNR numbers that can be compared to others codecs, e.g. from DivX and
> > XviD or ffmpeg.
> 
> The version of VP3 that I have includes a de-noise filter prior to 
> encoding. I am too unskilled to be able to edit the source to remove 
> this.

That's okay, if you still measure PSNR of input vs. output and not PSNR of
denoised vs. output. Every codecs should be free to do as much internal
preprocessing as it wants. That's part of the "intelligence". But the
results should be benchmarked against the original, of course. 

> > If there isn't a possibility to encode raw YUV or RGB with quicktime
> > codecs, you really should create one. I can't think of developing a 
> > codec
> > without it.
> 
> QT can give YUV, or RGB data to codecs, depending on how they declare 
> what colourspaces they accept. 

No, I meant: to encode from a raw YUV stream, just uncompressed YUV data
frame by frame without a MOV or AVI header or anything. 

> (3ivx, for example, has advertised a 
> great deal, that it can accept YUV video directly)

Wow! 
DivXnetworks has advertised that DivX will be faster on Opteron because of
"unrolled loops". 

Christoph 

--- >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 'theora-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.