[Feat req] Nerf->SDF->Mesh pipeline for AIGC

[yhyu13]

刚接触AIGC，有计划实现类似instant-ngp的网格体生成功能吗，只要白膜就行。这样就可以用于AIGC了

如果我没理解错instant-ngp的实现方法 https://github.com/NVlabs/instant-ngp/blob/master/src/marching_cubes.cu
原理应该是nerf 按照不透明度 转化成sdf，然后sdf再采样成mesh。

不确定太极有没有 现成的sdf2mesh函数库。但是找到一个开源的python实现：https://github.com/cvlab-epfl/voxel2mesh

谢谢

---

I am new to AIGC, are there any plans to implement mesh generation functions similar to instant-ngp, as long as the white film will do? This makes it useful for AIGC

If I'm not mistaken, instant-ngp implementation https://github.com/NVlabs/instant-ngp/blob/master/src/marching_cubes.cu
The principle should be that nerf is converted into SDF according to opacity, and then SDF is sampled into mesh.

Not sure if Tai Chi has a ready-made sdf2mesh library. But find an open source Python implementation: https://github.com/cvlab-epfl/voxel2mesh

[Linyou]

Hi @yhyu13

Thanks for bringing this up! Regarding your question, I currently do not have any plan to support mesh extraction.

However, you could consider using another repo like nerfmesh, which already offers a python implementation for your solution. And if you are interested in SDF, I suggest you could use sdfstudio