DDN-SLAM: Real-time Dense Dynamic Neural Implicit SLAM

Introduction

The paper is available in https://doi.org/10.1109/LRA.2025.3546130

DDN-SLAM is implemented with three open source projects

Orbeez-SLAM (GPLv3 license)
ORB-SLAM2 (GPLv3 license)
instant-ngp (Nvidia Source Code License-NC)

License

This repo is GPLv3 Licensed.

How to build

Install yolo_ros
Please refer to BUILD.md

RUN on TUM

Please refer to TUM.md

RUN on Replica

Please refer to Replica.md

RUN on ScanNet

Please refer to ScanNet.md

RUN on a custom dataset

Please refer to Custom.md

Evaluation

Please refer to Eval.md

Acknowledgments

This work has been possible thanks to the open-source code from Orbeez-SLAM,ORB-SLAM2 and instant-ngp, as well as the open-source datasets Replica and TUM-dataset.

Citation

If you find our work useful, please kindly cite us:

@ARTICLE{10904342,
  author={Li, Mingrui and Guo, Zhetao and Deng, Tianchen and Zhou, Yiming and Ren, Yuxiang and Wang, Hongyu},
  journal={IEEE Robotics and Automation Letters}, 
  title={DDN-SLAM: Real Time Dense Dynamic Neural Implicit SLAM}, 
  year={2025},
  volume={10},
  number={5},
  pages={4300-4307},
  keywords={Simultaneous localization and mapping;Semantics;Rendering (computer graphics);Neural radiance field;Feature extraction;Dynamics;Real-time systems;Optical flow;Accuracy;Interference;Dense visual SLAM;dynamic SLAM;neural implicit mapping},
  doi={10.1109/LRA.2025.3546130}}

DDN-SLAM