Just stumbled upon this one:
http://www.vrvis.at/br1/billboardclouds/
* Matzon likes the http://www.vrvis.at/br1/billboardclouds/fuhrmann_extreme_model_simplification_1.5_hi.mpg
Just stumbled upon this one:
http://www.vrvis.at/br1/billboardclouds/
* Matzon likes the http://www.vrvis.at/br1/billboardclouds/fuhrmann_extreme_model_simplification_1.5_hi.mpg
That is awesome… if there were a bit more random variation in the trees it would look really nice. I think Far Cry could have used that technology
That is very swav!
Perhaps we should make a tool for the java community?
Nice!!
this as a tool for the java community would be great ;D ;D
Ok…who wants to start making one then? :
Darn impressive!
Has anyone read enough to grasp how they’re choosing the billboards to generate yet?
Kev
As far as I understood it, they are using something like a Point-cloud (maybe better: face-cloud) from the original big tree-data and then they find billboards that match good for many points. The epsilon value specifies what a “good match” is. Then, when the Planes are calculated they create the textures by projecting the high-poly tree onto the billboards.
So that’s also the reason, why most of the trees look similar, because they are not generated on the fly, but from existing high-poly trees.
Yeah, the interesting bit is the billboard choosing bit though, just picking the rectangles that have the highest intersect isn’t always going to work well… they easily randomise it a bit to get rid of the same looking trees but I’m interested to know what the heuristic for picking the boards is. It could be as you say just the highest intersect.
Kev
I’ve read somewhere that they are using some sort of greedy heuristic.
maybe somebody could simply send them an email asking for their code, so we can port it?