I had the pleasure, in the past two days, to participate at a workshop called Fokus3D. It was the closing event of a European R&D project of a similar name, concentrating on what is called Semantic 3D. I was invited because the project made use of certain types of Semantic Web technologies (e.g., OWL) and, also, because it is the community of my previous professional life: I did spent many years in Computer Graphics… (Which also meant that I met old friends and colleagues that I had not seen for many years, which was really very pleasant…)
So what is this “Semantic 3D”? What does it have to do with Semantic Web?
Here is a a crash course on 3D graphics: when systems display those beautiful graphics 3D objects that we are all used to, the underlying system transforms complex mathematical descriptions of shapes, surfaces, or 3D bodies into a load of (triangular) meshes that are displayed by the graphics hardware. The mathematical descriptions are purely geometrical and define, say, spline surfaces, planes, or some geometric transformations that place those surface description into space.
These 3D objects represent, usually, some real object. A chair, a car, a tree, or a house. The representation of a chair is a combination of several such shapes; some of those describe the arms, the back, etc. But this information, i.e., that this and this combination of shapes is actually the arm of a chair, is usually lost somewhere in the process. Modelers start with a concept, a “semantic”, and end up with shapes; information is gone on the way. This means that many things cannot be done well: one would like to have semantics based search (“searching for the arm of a chair”), one would like to know the origin of a a particular shape (i.e., how was it created, under what process and transformation), one would like to follow the evolution in time of a particular shape to retrace the designer’s actions, etc., etc. And, due to the huge number of shapes, managing this type (meta) data is far from obvious. Keeping that information in a manageable way together with the geometric processing: we get Semantic 3D.
There was, of course, a slight confusion of terms for me: this notion of semantics would be considered as (meta)data for Semantic Web people. That being said, such data requires controlled vocabularies, and very complicated ones at that, so there are strong connections nevertheless. But there is also semantics in terms of knowledge representation: There are relationships among, and classification of, shape elements, these relationships can represent constraints and other features that can be used for reasoning, for inference. So more complex ontologies come into the picture (and OWL is widely used in this space). These ontologies are often application dependent, reflecting the diversity of application areas from CAD to gaming, or from cultural heritage systems to medical and biological applications. In future, such ontologies should also incorporate features like uncertainty (reflecting the fact that, at least in some areas like protein modeling, those relationships are not necessarily crisp); they should also include features such as provenance or time relationships.
Last but not least: there are lots of data there. I mean lots, stored in biological databases, shape libraries, scanned historical artifacts, each representing an object (like the reproduction of the Ramses statue on the figure) with many many shapes. Integration of that data is a challenge even within one application area, let alone with data at large. It will take a long time when this data will be organized in a way that it could be, say, exposed and integrated as Linked Open Data. But we may get there, eventually (and your truly has done his best to convince the community of the value of doing that…). Standard representations have to be developed, algorithms crystallized, vocabularies and ontologies defined, etc. The good news is that there is a community that is determined to continue working in this direction. The workshop organizers plan to write down a research roadmap (to be put on line within 1-2 weeks), and a special issue of the journal “Computer & Graphics” has been announced, co-edited by Bianca Falcidieno, from the CNR in Genova, and myself. So… stay tuned.