1.Installation

August 8, 2025 · View on GitHub

4D Gaussian Splatting SLAM

Yanyan Li · Youxu Fang · Zunjie Zhu · Kunyi Li · Yong Ding · Federico Tombari

Paper | Project Page

1.Installation

git clone https://github.com/yanyan-li/4DGS-SLAM.git
cd 4DGS-SLAM

Setup the environment.

conda create -n 4dgs-slam python=3.8
conda activate 4dgs-slam
# CUDA 11.7
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
pip install -r requirements.txt

The simple-knn and diff-gaussian-rasterization libraries use the ones provided by MonoGS.

pip install submodules/simple-knn
pip install submodules/diff-gaussian-rasterization

Use torch-batch-svd speed up (Optional)

git clone https://github.com/KinglittleQ/torch-batch-svd
cd torch-batch-svd
python setup.py install

2.Pretrained Models

Download YOLOv9e-seg

cd 4DGS-SLAM/pretrained
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9e-seg.pt

Or download it directly from https://docs.ultralytics.com/models/yolov9/

Download RAFT

The model raft-things.pth used in this system can be obtained directly from https://drive.google.com/drive/folders/1sWDsfuZ3Up38EUQt7-JDTT1HcGHuJgvT

3.Datasets

TUM-RGBD dataset

Download the sequence using the following command:

bash scripts/download_tum_dynamic.sh

4.Testing

Dynamic rendering

python slam.py --config configs/rgbd/tum/fr3_sitting.yaml --eval --dynamic

Adjust the frequency for image saving

python slam.py --config configs/rgbd/tum/fr3_sitting.yaml --eval --dynamic --interval 50

5.Acknowledgement

This work incorporates many open-source codes. We extend our gratitude to the authors of the software.

6.Citation

If you find this code/work useful for your own research, please consider citing:

@article{li20254d,
  title={4{D} {G}aussian {S}platting {SLAM}},
  author={Li, Yanyan and Fang, Youxu and Zhu, Zunjie and Li, Kunyi and Ding, Yong and Tombari, Federico},
  journal={arXiv preprint arXiv:2503.16710},
  year={2025}
}