ERP said:
I think the largest valence in the model was 8 but this was close to 5 years ago and I'd have to look at it to verify. By far the most common valence (other than 4) was 5.
Pardon me for being rude, but with such high valences, you should get another artist
A vertex with 8 edges is totally unneccessary, there are modeling strategies to deal with such cases.
It's actually a somewhat complex and often argued problem in the online CG community, which has moved to Catmull-Clark subdivs almost exclusively in the past years. Basicaly, you can have n-sided polygons and n-sided vertices (where n is the number of adjacent edges meeting at that vertex). You can build a polymesh that has either
- only 4-sided polygons (quads) and n-sided vertices, or
- only 4-sided vertices and n-sided polygons.
Generally, the first version is prefered, because it gives more control over the surface curvature for the artist. It's because an n-sided polygon gets subdivided to n quad polygons, with an n-sided vertex in the middle, but there's no direct control over that vertex. Also, the topology will tend to have a spiderweb-like structure at this vertex, which gets worse and worse as the number of edges increases - an 8-sided vertex is a truly horrible thing and will most likely cause some ugly pinching on the subdivided surface.
The problems with maintaning an all-quad polymesh, with only a few 3 and 5-sided vertices are that it takes a lot more time to build, and it is generally more dense than a less organized mesh with n-sided polygons. So there are supporters for both approaches, and neither has been declared a winner yet
Hard surfaces like cars, weapons, armor etc. are even more complicated to build from quad polys only, partially because surface smoothness is a lot more important here compared to characters. However I've seen a lot of very nice car models that looked prefectly smooth in high-res images with good AA/filtering (that would bring out any surface problems). In the end, complex NURBS surfaces are the best way to model them, but they have a huge computation hit for all the procedurally performed blends, fillets and trims to get those nice shapes we see on today's cars. But NURBS modeling is very time consuming as well, so modelers tend to stick with polygons.
The problem here is what ERP has already mentioned somewhere - to get enough control over the subdivided surface, you'll usually have to increase the density of the control mesh to a level where it's usually more efficient (at this time) to skip subdivision at all. This will change in the next few years however, at least IMHO.
Another, more generic problem with tesselation is that you can (and will) waste a lot of processing time and memory for objects far away from the camera. Even in the controlled enviroment of CG, it's always better to have both lazy and adaptive tesselation to keep render times manageable. Thus it is especially important in a realtime enviroment to only perform tesselation when it's worth doing. But AFAIK all of today's hw tesselation implementions, with the exception of the Xenos, are lacking any support for adaptive/lazy tesselation, and I believe that it's the reason why they aren't used at all.
They should be free to use whatever tool they like then they should provide a dense mesh.
This mesh must be (automatically) remeshed to fit with proprietary 3D engine costraints (you can't support any vertex valence, any kind of topology, and so on..).
The difficult part is the remeshing phase, without this phase is a comlpetely waste of time to even try to support SS in a realtime engine.
I dunno, but I imagine that any procedural remeshing would produce results inferior to what a skilled modeler can deliver. I've never heard of any remeshing that could build a good face or hand, for example... keep in mind that you'd also have to weight this newly created mesh for skeletial animation, where it's increasingly important to have good topology.