search

nerfstudio-project / nerfacc

A General NeRF Acceleration Toolbox in PyTorch.
https://www.nerfacc.com/
Other
1.38k stars 112 forks source link

Density threshold #121

Closed MaximKuklin closed 1 year ago

MaximKuklin commented 1 year ago

Hello! Thank you for your implementation! Right now I'm working on converting output density to PNG slices like it's done in the instant-ngp repo - link. These slices are used to generate a point cloud.

In instant-ngp they have a threshold=2.5 as default, so the final equation for png slices looks like slices=(density - 2.5) * 32 + 128.5 and then slices are clipped into the range from 0 to 255. So I'm curious about this 2.5 threshold and how to properly calculate it in your implementation. Do you have any tips on how to threshold density correctly?

liruilong940607 commented 1 year ago

Hi you can still use 2.5 in nerfacc. I'm not sure if there is a way to properly calculate this threshold as it relates to your network and ray marching step size. Generally 2 ~ 5 works fine.

MaximKuklin commented 1 year ago

Thank you!