A real-time Gaussian Splatting implementation for SLAM systems, building upon the excellent work of MonoGS and Photo-SLAM.

git clone https://github.com/Nemo0412/MonoRTGS.git
cd MonoRTGS

Environment Note

• environment.yml is for x86 architecture.
• environment_arch.yml is for arm64 architecture (e.g., ONX).

conda env create -f environment.yml
conda activate MonoRTGS

The following scripts will automatically download datasets into the ./data folder.

bash scripts/download_tum.sh

bash scripts/download_replica.sh

This project includes convenient scripts to run both the original version (MonoGS) and RTGS version:

python3 run_slam.py <version> <config>

./run_slam.sh <version> <config>

# Run original version
python3 run_slam.py original tum/fr3_office
./run_slam.sh original tum/fr3_office
# Run RTGS version
python3 run_slam.py RTGS tum/fr3_office
./run_slam.sh RTGS tum/fr3_office

• tum/fr1_desk
• tum/fr2_xyz
• tum/fr3_office

• replica/office0
• replica/office1
• replica/office2
• replica/office3
• replica/office4
• replica/room0
• replica/room1
• replica/room2

RTGS/
├── run_slam.py              # Python version SLAM runner script
├── run_slam.sh              # Bash version SLAM runner script
├── MonoGS/                  # Original version
├── MonoGS_RTGS/             # RTGS version
├── configs/                 # Configuration files
├── scripts/                 # Download scripts
└── data/                    # Dataset directory

You can also run this project directly using Docker, without manual environment setup.

docker pull mugen0412/monortgs:cuda12.1
docker run --rm -it --gpus all mugen0412/monortgs:cuda12.1 bash

For a RTGS-SLAM implementation based on Photo-SLAM, please check out:

• PhotoSLAM RTGS Implementation

This project builds upon the excellent work of the authors of MonoGS and Photo-SLAM.
We gratefully acknowledge their open-source contributions, which make this project possible.

• MonoGS (CVPR 2024)
• Photo-SLAM (CVPR 2024)