|prev : next : index||SPEW|
Trees are some of the hardest things for computers to draw. There are a couple of reasons for this. One is that they are basically made up of a lot of little discrete elements (leaves). Humans may be made up of lots of discrete elements (molecules), but they join together into much larger continuous features, which can usually be represented by a simple drawing primative.
A low-quality obviously-blocky human can be drawn in 100 polygons; a decent one in 1000. One that looks good in close-ups requires a lot more.
A tree requires tens of thousands of polygons to represent all the branches and leaves. And there aren't really any good shortcuts. This is a problem when you need hundreds of trees for a scene to be convincing.
Since that would basically mean no trees, people have hacked around it with some particularly fake looking things, that look like trees from far away, but become a pretty obvious approximation up close.
I'm working on a demo of a novel detail-reduction technology that allows up-close trees to be drawn as fully detailed models, and uses a similar tactic as the above hack to approximate them when far away. The trick here is that the system can smoothly convert between the two versions, so your eye won't see it switch. We call this continuous level-of-detail in the business. Active research (e.g. in Siggraph) has focused on continuous level-of-detail for continuous surfaces, in which a set of adjacent continuous surfaces (e.g. polygons) can be drawn as a single surface (polygon) when the object is small. So this research, which allows discontinuous surfaces to be so drawn, is actually pretty novel.
It doesn't really work as well as I'd like, but it's actually a real step in a direction nobody seems to have seriously taken before.
No, most of my journal entries aren't going to be this techy.