fatPeter / mini-splatting

92 stars 9 forks source link

Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians (ECCV 2024)

This is the official implementation of Mini-Splatting, a point cloud analysis work in the context of Gaussian Splatting. Through reorganizing the spatial distribution of 3D Gaussians, our algorithm improves the model performance without any auxiliary information. For technical details, please refer to:

Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians
Guangchi Fang and Bing Wang.
[Paper]

(1) Setup

This code has been tested with Python 3.7, torch 1.12.1, CUDA 11.6.

(2) Mini-Splatting (Densification and Simplification)

Training scripts for Mini-Splatting are in ms:

cd ms

(3) Mini-Splatting-D (Densification only)

(4) Mini-Splatting-C (Mini-Splatting with Compression)

Acknowledgement. This project is built upon 3DGS.