Page 1 of 1
High Poly test
Posted: Sun Apr 15, 2007 8:54 pm
by OnoSendai
Just a little test render, using the bounding interval hierarchy tree type (BIH).
Model is from here:
http://graphics.stanford.edu/data/3Dscanrep/
Each dragon has 7.2 million tris.
I'm using instancing to repeat the model.
Total Indigo mem usage is about 730MB.
Render time is ~20 mins
Posted: Sun Apr 15, 2007 9:24 pm
by suvakas
whoo...this is nice !!!
Posted: Sun Apr 15, 2007 10:34 pm
by zuegs
Thats cool!!!
Just a small question: how you do instancing with changing materials? As i remember the material is assigned inside the <mesh>. Is there a way to override that inside the <object> (in case that mesh is declared in XML, not external 3ds-file) ???
Posted: Sun Apr 15, 2007 10:55 pm
by Big Fan
edited to remove all
Posted: Sun Apr 15, 2007 11:05 pm
by OnoSendai
Wow nice render Big Fan, blender internal is actually pretty good
Posted: Sun Apr 15, 2007 11:15 pm
by OnoSendai
zuegs wrote:Thats cool!!!
Just a small question: how you do instancing with changing materials? As i remember the material is assigned inside the <mesh>. Is there a way to override that inside the <object> (in case that mesh is declared in XML, not external 3ds-file) ???
define stuff in this order:
mesh
material
model
material
model
material
model
Posted: Mon Apr 16, 2007 8:38 pm
by mrCarnivore
7.2 million tris? not bad in 20mins!
I guess when using sss, this render will take like 2 months or something...
Posted: Mon Apr 23, 2007 10:58 am
by IanT
Very nice Nick ... so how are you finding BIH vs. KD-Tree? I'm currently messing about with nested grids at the moment ... seems to outperform KD-Tree (but only just) but with about a fifth of the memory requirements.
(Sorry to go Off-topic
)
Ian.
Posted: Mon Apr 23, 2007 11:34 am
by dogfin
Just for fun I did some reading on both algorithms, but I couldn't find the answer to one thing. How well do they perform at the "teapot in a stadium" problem?
dogfin -
Posted: Mon Apr 23, 2007 12:15 pm
by IanT
dogfin wrote:Just for fun I did some reading on both algorithms, but I couldn't find the answer to one thing. How well do they perform at the "teapot in a stadium" problem?
dogfin -
They're both pretty effective, but the BIH will generally use significantly less memory than KD-Tree.
Ian.
Posted: Mon Apr 23, 2007 11:02 pm
by OnoSendai
IanT wrote:Very nice Nick ... so how are you finding BIH vs. KD-Tree? I'm currently messing about with nested grids at the moment ... seems to outperform KD-Tree (but only just) but with about a fifth of the memory requirements.
(Sorry to go Off-topic
)
Ian.
BIHs are nice to work with. The build times are super fast, and the mem requirements are very low, so for large meshes, they are superior to kd-trees.
In terms of ray tracing performance, my BIH is getting about 30-90% of the kd-tree performance.
It's early days however, and I haven't made any careful measurements, just looking at the Mutations/s
If you're outperforming kdtrees with your nested grids, that's quite an achievement (unless your kd-tree implementation sux
).
I'm presuming your nested grids can cope with the teapot/stadium issue which plagues uniform grids?