This page provides an exprimental web viewer for Octree-GS [Ren 2024], which is a state-of-the-art extension of Gaussian-Splatting [Kerbl 2023]. You can try online demo form here. The code is based on Gaussian-Splatting-WebViewers (gaussian_splatting_1), which is originally based on WebGL implementation of Gaussian Splatting.
npm install
npx webpack
npx light-server -s . -p 8080
and open your browser with http://localhost:8080/
-
Place the trained data of Octree-GS as follows
root/
├── gs-data/
│ ├── point_cloud.ply
│ ├── color_mlp.pt
│ ├── cov_mlp.pt
│ ├── opacity_mlp.pt -
Set up python environment
conda env create --file environment.yml
conda activate octree- -
Click on “run.bat” to convert the data
The current version is slow in rendering speed, so any advice for speeding it up is welcome!
Scaffold-GS precedes Octree-GS; both were published by the same research group and have the same data format. Therefore, this viewer may be applicable to Scaffold-GS, but we have not tried it.
convert-to-stabdard-ply.py can convert octree-gs fromat to standard-gs format by assuming a fixed camera position.