[Theora] Ogg Theora: Some questions on codec interna

[Ralph Giles]

On Mon, Sep 20, 2004 at 11:43:35AM +0200, Gernot Ziegler wrote:

> Actually I like the codec's style of omitting interlaced handling and
> B-frames, makes the format much more simple - but how do you handle
> interlaced material then, and can you achieve the same compression as MPEG
> even without B-frames ?

Yes, it is a nice simplification.

According to Dan Miller, who worked on VP3 at On2, they didn't need them 
to match MPEG. Instead theora can refer to the previous I frame as well 
as the previous frame. We could perhaps do better with B frames, but not 
enough better to warrent the complexity.

We deal with interlaced material by de-interlacing it before encoding. 
This a general philosophy with both theora and vorbis: keep the decoder 
simple, but flexible, and put shift as much intelligence and 
decision-making to the encoder as possible.

Generally, for playback on any progressive-capable device, this is what 
you want anyway. Playback to standard televisions (set-top box 
applications) is really the only place being able to directly encode 
interlaced material makes any sense.

> > So, you have a range of options, but it's unlikely to do better than
> > mjpeg in terms of bitrate unless a large part of your sequence is
> > static or a simple pan. But no problems with deep image support.
> ("deep image support" = " 24 bit per pixel" ? )

"deep images support' = "more than 8 bits per sample", up to 64 bits per 
pixel in the case of MNG (RGBA).

> M-JPEG is no option as it would also compress the depth data with JPEG,
> which is DCT-based, which is bad ;)

Hmm. Right. I wonder if you could experiment with 'fixing up' the jpeg 
depth with additional composites? Might not be too effective in terms of 
bitrate though.

Cheers,
 -r