[vos-d] VOS requirements

[Karsten Otto]

Am 25.01.2007 um 16:27 schrieb Len Bullard:

> X3D has a physics specification underway.   One is already being  
> integrated
> into Contact.  X3D already has shaders and scripting plus a  
> metadata node
> for indicating semantics.  Since the objects you mention below can  
> be notatd
> as say DEF Tree and referenced by that name, I'm not sure what you  
> want for
> semantics past that which won't create a badly layered design.
>
> Collada was designed as a transfer format for games.  It is  
> compatible with
> X3D.
>

Ok, I see this is a difficult concept to get across when people are  
used to scene graphs. Let me try to explain it another way. Yes, you  
can group primitives via DEF into a complex shape, and even give it  
some name that suggests to the human reader what is meant by this  
group. The problem is that DEF in VRML is designed as a syntactic  
tool, not as a signifier for complex objects. Remember that you could  
as easily group part of a world for editing convenience.

world --> transform --> DEF tree = (transform1--> sphere, transform2  
--> cylinder) --> meta: "tree"

and

world --> transform1 --> sphere --> meta: "treetop"
world --> transform2 --> cylinder --> meta: "treetrunk"

will be both be percieved by a human as a tree, but for an agent both  
cases are still gibberish. The label "tree" does not help either, it  
could as well have been "Baum" or "Arbol"; these are words that  
require understandig of a human language to process, and agents  
generally are not capable of that. You need a controlled vocabulary,  
semantic URI, or similar concept identifier that is machine  
understandable.

Finally, meta nodes may help a bit, but for an agent to userstand the  
scene it would first have to extract them all, pick the semantically  
relevant parts of the scene graph, and correlate them somehow. In the  
first case this is easier than in the second case; sadly 3D modelling  
tools tend towards the second case, especially when naive users toy  
with them.  After all, it "looks" great, right? Unfortunately its a  
bad situation for an agent interested primarily in the meta information.

I think this boils down to intended use. If you want a scene looking  
good to a human, use a scene graph. If you want a machine  
understandable scene, use a description language. If you want *both*,  
well... pick a focus, and either emphasize the scene with attached  
meta nodes, or emphasize the semantic description and attach the  
geometry. Scene emphasis is mainstream, I happen to prefer semantic  
emphasis. Think outward and inner beauty :-)

Regards,
Karsten Otto (kao)