Informative Scene-Reconstruction App

LLM Scene Reconstructor V0.0.1

A local software and cloud service system that integrates 3D functionalities including sparse reconstruction, point cloud regions annotation, 3D Gaussian Splatting (3DGS) real-scene model construction on-cloud, real-scene rendering on-cloud, and LLM-agent guidance. The system can support applications in tourism and disaster overview, and currently supports fast deployment on both Local (Windows 11) and Server (Linux) environments.

Haojun Tang, Jingran Zhang, Siyuan Zou

Central South University

🎞️ Demonstration

Demo Video (in Chinese)

Youtube Video Link

🏠 Sturcture

System Structure:

💻 Deployment

Environments

Server:
CUDA = 11.8

conda create -n scenereconstruction python=3.9
conda activate scenereconstruction

Local (tested on Windows 11):
CUDA = 12.4

conda create -n scenereconstruction python=3.9
conda activate scenereconstruction

Dependencies and Running

Server:

pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements_server.txt
uvicorn server.main:app --host 0.0.0.0 --port 8000 (Port 8000 on Server needs to be exposed to Local)

Local:

pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements_desktop.txt
python main.py

📒 To-do List

• Core Pipelines
• Progress Dialogs
• English Version
• Reconstruction from Videos
• Search Engine for LLM-Agent
• Multi-GPUs Support

🤝 Acknowledgements

Thanks to: PyQt5, PyColmap, PyOpenGL, PyVista, OpenAI Python, Open3D, WebSocket-Client, FastAPI, uvicorn, plyfile, 3DGS, Python-Multipart

📝 License and Citation

License

This project is released under a Non-Commercial Research License.
The software is free for academic research and non-commercial use.
Commercial use requires a separate license from the authors.
See the LICENSE file for details.
For commercial licensing, please contact:
rs_lover@163.com

Citation

If you find this project useful in your research, please consider citing:

@misc{tang2026informative,
    title = {Informative Scene-Reconstruction App},
    author = {Haojun Tang and Jingran Zhang and Siyuan Zou},
    year = {2026},
    note={GitHub repository},
    howpublished={https://github.com/DonaldTrump-coder/Informative-Scene-Reconstruction-App}
    }