search

sxyu / svox2

Plenoxels: Radiance Fields without Neural Networks
BSD 2-Clause "Simplified" License
2.79k stars 358 forks source link

Compare to DirectVoxGO #7

Closed BrianPugh closed 2 years ago

BrianPugh commented 2 years ago

I just noticed a new, fairly similar method to yours, DirectVoxGo. I was wondering if you could perform some comparisons/comments on their technique vs yours? Thanks!

sxyu commented 2 years ago

Hi Brian, thanks for the suggestion. DirectVoxGo is an excellent work and I have been in contact with the author. Actually the website https://alexyu.net/plenoxels/ contains a section on this work with some comments on the differences. Mainly (1) they are still using an MLP (2) they do not use any CUDA (only pure PyTorch) (3) they rely heavily on tightening the bounding box to get higher resolution (in our case, we simply sparsify the grid)

BrianPugh commented 2 years ago

thank you!

incase you haven't seen it, there's another paper "PERF: Performant, Explicit Radiance Fields" that doesn't use neural networks. The authors didn't release any code, but an interesting and relevant read

https://arxiv.org/pdf/2112.05598.pdf