MVS-GS / Online 3D Gaussian Splatting with Novel View Selection

🚀 Official repository for:

• Online 3D Gaussian Splatting Modeling with Novel View Selection (IJCAI 2025)
• MVS-GS: High-Quality 3D Gaussian Splatting Mapping via Online Multi-View Stereo (IEEE Access 2025)

👉 Paper (arXiv)

---

🔥 Overview

We present an online 3D Gaussian Splatting framework that reconstructs a scene from an RGB image stream.

The system integrates SLAM, multi-view stereo, and neural rendering into a unified pipeline:

• Informative view selection for efficient reconstruction
• Multi-view stereo for geometric consistency
• Online mapping and optimization

---

🧩 Pipeline

The framework consists of three main components:

1. Tracking / Pose Estimation Camera poses are estimated from RGB inputs using DROID-SLAM

2. Online Depth Estimation Multi-view consistent depth is predicted using MVSFormer

3. Mapping / Optimization Online 3D Gaussian Splatting with:

   • MVS-based geometric initialization
   • Novel view selection for efficient updates

---

⚙️ Installation

1) Create environment

conda create -n mvs_droidslam python=3.8 -y
conda activate mvs_droidslam

conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 \
  -c pytorch -c nvidia

---

2) Clone repository

git clone --recursive https://github.com/lbg030/MVS-Droid_SLAM.git
cd MVS-Droid_SLAM

---

3) Install dependencies

pip install -r requirements.txt

---

4) Build CUDA extensions

python setup.py install
pip install ./submodules/diff-gaussian-rasterization
pip install ./submodules/simple-knn

---

📦 Pretrained Weights

DROID-SLAM

pip install gdown
mkdir -p weights
gdown --fuzzy "https://drive.google.com/file/d/1PpqVt1H4maBa_GbPJp4NwxRsd9jk-elh/view?usp=sharing" \
  -O weights/droid.pth

---

MVSFormer

Download from: https://onedrive.live.com/...

mkdir -p weights
# place checkpoint at:
# weights/mvsformer.pth

---

🚀 Quickstart

python demo.py --config <path_to_config.yaml>

---

📂 Outputs

The system produces:

• Reconstructed 3D Gaussian model
• Camera trajectory
• Depth and intermediate mapping results

---

📄 Citation

If you find our work useful, please consider citing:

MVS-GS (IEEE Access)

@article{lee2025mvs,
  title={Mvs-gs: High-quality 3d gaussian splatting mapping via online multi-view stereo},
  author={Lee, Byeonggwon and Park, Junkyu and Giang, Khang Truong and Jo, Sungho and Song, Soohwan},
  journal={IEEE Access},
  year={2025},
  publisher={IEEE}
}

---

Online 3D Gaussian Splatting with Novel View Selection (IJCAI2025)

@article{lee2025online,
  title={Online 3D Gaussian splatting modeling with novel view selection},
  author={Lee, Byeonggwon and Park, Junkyu and Giang, Khang Truong and Song, Soohwan},
  journal={arXiv preprint arXiv:2508.14014},
  year={2025}
}

---

📜 License

See LICENSE