Skip to content

Latest commit

 

History

History
41 lines (21 loc) · 2.14 KB

README.md

File metadata and controls

41 lines (21 loc) · 2.14 KB

High-fidelity 3D mesh reconstruction using Gaussian Splatting (GS): Research and Evaluation

This repo contains utility and visualization scripts related to the mesh reconstruction based on Gaussian splatting.

image

Methods

SuGaR

For SuGaR, there is a bash script for training a Gaussian splatting model

Please clone the repository of dn-splatter, and follow the installation instructions. The training script should be placed in the root directory of SuGaR.

Under notebooks, you can find a Jupyter notebook for SuGaR visualization and evaluation.

dn-splatter

For dn-splatter, there are bash scripts for training a Gaussian splatting model and mesh extraction & evaluation.

Please clone the repository of dn-splatter, and follow the installation instructions. The scripts should be placed under the root directory of dn-splatter.

I would recommend applying a patch to avoid segfault when extracting mesh using gaussians method.

Evaluation

A Python evaluation script can be used to calculate two variants of Normalized Chamfer distance between a ground truth and reconstructed mesh.

It should be placed in the parent directory of dn-splatter and SuGaR.

Misc

SAM2

Under notebooks, you can find a Jupyter notebook for object mask prediction using SAM2.