dazinovic / neural-rgbd-surface-reconstruction

Official implementation of the CVPR 2022 Paper "Neural RGB-D Surface Reconstruction"

Home Page:https://dazinovic.github.io/neural-rgbd-surface-reconstruction/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

use extract_mesh.py to extract mesh from model trained on scannet

XuqianRen opened this issue · comments

Hi, thank you for your project. I have one question. When I try to use the extract_mesh.py script in your project to extract mesh for scannet scenes, the mesh results are all in a mess as followings, some objects have been reconstructed, some are not, and they are all very noisy.
Like for scene0005:
image

scene0050:
image

But I have got the right nerf result for scannet scenes: (like for scene0050)
image

Also, when running this script with the synthesis dataset, the results are clear.
image

I want to know is there any change I need to add to adapt this script to fit the scannet dataset? Or is there any problem in my way to training with scannet dataset?

commented

Hi Xuqian,

It looks like the region from which you are trying to extract the mesh is incorrect. The scene is repeating because of the periodic positional encoding. However, only the region that was optimized will look good. The quality of the repetitions is always degraded.

Thank you