Motivation
This repository aims to bridge the gap between academic research and industrial deployment of 3D Gaussian Splatting (3DGS) technology. While most existing papers focus on theoretical innovations, this project emphasizes:
- Engineering-oriented Implementation - Providing practical, scalable 3DGS solutions with a focus on performance and stability in real-world application scenarios.
- Steady Performance Optimization - Systematically improving and optimizing 3DGS to enhance operational efficiency across various hardware configurations and environments.
- Systematic Documentation of Practical Workflows - Comprehensive documentation of the complete pipeline from data acquisition and camera pose estimation to 3DGS training and application.
Our project is built upon the original 3DGS.
Dataset
| Dataset | Model | PSNR | SSIM |
|---|---|---|---|
| KITTI | 3DGS_base | 28.50 | ---- |
| ABS-GS | 28.55 | ---- | |
| RAIN-GS | ---- | ---- | |
| smith_hall_outdoor | 3DGS_base | 24.85 | ---- |
| ABS-GS | 24.64 | ---- | |
| RAIN-GS | ---- | ---- | |
Qualitative Results
Workflow
Stage_0. Data Acquisition
Stage_1. Camera Pose Estimation
Stage_2. 3DGS Training
Stage_3. Export & Deployment
Bibtex
@inproceedings{li2025scenesplat,
title={gaussian-splatting-engineering},
author={EvangelionA},
year={2025}
}
Acknowledgement
This project is built upon 3DGS and This Project-webpage is built upon SceneSplat We thank all the authors for their great repos.