zju3dv / Vox-Fusion

Code for "Dense Tracking and Mapping with Voxel-based Neural Implicit Representation", ISMAR 2022

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The net setup is not same as the article described

JunyuanDeng opened this issue · comments

I notice that in the file "replica.yaml", you set depth: 2 and width: 128. I checked in src/variation/nrgbd.py, these means the 2 FC layers for sdf and 128 hidden units. But in your article, there are 4 FC and 258 hidden units. Why there are differences between real implementation and article description?

Thanks for your reply!

Thanks for pointing it out.

We did some experiments with fewer network parameters recently and found similar results. However, you can revert to the same parameters as the article.

Thanks for your reply, now I understand!