I came across a site today talking about a GPU-friendly terrain rendering algorithm for large datasets. The method described in the paper looked pretty simple, so I hacked together a quick implementation.
That turned out to work pretty nice, so I dusted off some old procedural planet texture generation stuff I worked on a while back and started to combine the two. While I didn’t get to the point of doing textures, I did manage to map the LOD system to a sphere. It works pretty nicely – moving in close ramps up the detail, zooming out degrades the mesh to almost nothing.
Here’s another shot not in wireframe mode. As you can see, I am yet to implement the crack-filling system that fixes the seams between parts of the mesh 
The textures I generated in the old system didn’t look too bad, if a little cartoony. With any luck it won’t take much to port it over to this new renderer – here are some old images, just for kicks.
[...] posted previously, I’ve been playing with the beginnings of a LOD terrain renderer for large datasets. I [...]
[...] than that, I’m quite pleased. It’s come a long way from a couple months back, considering I only hack on this as a break from more important work. « [...]