Here you can see the Standford dragon after voxelization in a 128x128x128 grid:
The original model has 260,000 triangles. The resulting mesh resolution for 128x128x128 is much lower, so a lot of detail is necessarily lost. Still the voxelization preserves the key features of the model.
This runs in 14 milliseconds on an ATI 4770.
What are thoughts on the practical limits of voxelization? In regards to, ideally you would want one voxel aligned to each texel on the input poly model. I've been musing with the idea of working back from the texels instead of axis aligned. Also have you considered a way of building voxels directly from L-Systems rather than the intermediate poly step?
ReplyDelete@Kerrash: I too think polygons will be eventually out and it will be voxels all the way. The problem right now is most art assets are already in polygonal form or will be created by polygonal edition tools like Maya or Max.
ReplyDeleteBut even then polygonal models will retain some core advantages. It is like vector graphics vs. bitmaps. For instance polygonal models are resolution independent, voxels will always show artifacts due to aliasing when resampled. For some applications, it will be better to store information as polygons.
I have a question, when you rotate the model, do u re-voxelize the model? since each small voxel is a cube which align with the X/Y/Z axis.
ReplyDeleteNo, he does not, whole model is voxelized.
Delete