I implemented a cuda version of converting mesh into o_voxel

[quantaji]

Hi, Thank you so much for this great research. I found that the o_voxel's implementation only use cpu to convert mesh into dual vertices. So I try to make a cuda version of this convertion and try to replicate the process as much as possible.

For the difference between gpu and cpu, I think the relative mse is 1e-5 im maxium. Please check out the evaluation notebooks in the o-voxel/notebooks folder. The speedup is around 10x, and may require less memory/vram.

Please let me know if you need more information.

Thank you agian for this great research!
Guangda

[quantaji]

@microsoft-github-policy-service agree

[quantaji]

Hi @yzg1, I pushed an updated faster version after your previous approval. Could you please take another quick look when you have time?

The main change is the updated o-voxel GPU implementation. I also added the standalone repo here:
https://github.com/quantaji/o-voxel-gpu

It looks like the CodeQL workflow may require maintainer action as well.

Thanks!