DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization

[Paper]

DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization,
Yueming Xu*, Haochen Jiang*, Zhongyang Xiao, Jianfeng Feng, Li Zhang
NeurIPS 2024

Official implementation of "DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization".

🛠️ Pipeline

⚙️ Installation

Please follow the instructions below to install the repo and dependencies.

git clone https://github.com/fudan-zvg/DG-SLAM.git
cd DG-SLAM

Install the environment

# Create conda environment
conda create -f environment.ymal
conda activate dgslam

If the front-end fail to execute successfully, please consult the official repository and accompanying guidance DROID-SLAM. The pretrained weights for DROID-SLAM can also be downloaded from this repository. The .path file should be placed in the ./checkpoints folder.

📂 Download Dataset & Data preprocessing

TUM RGB-D Dataset

Download 6 dynamic scene sequences of TUM RGB-D dataset into ./data/TUM folder.

bash scripts/download_tum.sh 

BONN Dynamic RGB-D Dataset

Download 6 dynamic scene sequences of BONN RGB-D dataset into ./data/BONN folder.

bash scripts/download_bonn.sh 

ScanNet Dataset

Please send an official email to request permission for dataset access.

Data preprocessing

• Semantic Segmentation Mask: Firstly, you should generate semantic motion mask with OneFormer. For guidance on the environmental setup and execution process for this approach, please refer to the instructions provided on the official GitHub page. Secondly, move these generate mask images to the root path of dataset and create the folder named ./seg_mask

🔄 Run

You can run DG-SLAM using the code below:

python run_tum.py

📜 BibTeX

If you find our code or paper useful for your research, please consider citing:

@inproceedings{xu2024dg,
  title={DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization},
  author={Xu, Yueming and Jiang, Haochen and Xiao, Zhongyang and Feng, Jianfeng and Zhang, Li},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2024}
}

Acknowledgement

Thanks to previous open-sourced repo: 3D-GS, Point-SLAM, DROID-SLAM, SplaTAM.

Contact

Contact Yueming Xu and Haochen Jiang for questions, comments and reporting bugs.