CP-SLAM: Collaborative Neural Point-based SLAM System[NeurIPS'23]

Jiarui Hu¹ · MaoMao¹ Hujun Bao¹ · Guofeng Zhang¹ · Zhaopeng Cui^1*
1 State Key Lab of CAD&CG, Zhejiang University
^* Corresponding author.

🌐Project page | 📝Paper | 📽️Video

This is the official implementation of CP-SLAM: Collaborative Neural Point-based SLAM System. CP-SLAM system demonstrates remarkable capabilities in multi-agent deployment and achieves state-of-the-art performance in tracking, map construction, and rendering.

Table of Contents

1. News
2. Dataset Download Link
3. Installation
4. Usage
   1. Run
   2. Evaluation
5. Acknowledgement
6. Citation

News

• 2024.04.11 --- We have updated the README.md and are preparing to open-source our code！
• 2024.04.26 --- Code for some functional modules, including loop detection, pose graph, federated center, and shared data structure (detailed comments will come soon).
• 2024.05.08 --- Code for main parts, including optimizer, renderer, fusion center, and tracking and mapping modules.
• Installation setup

Dataset Download Link

We provide the Download link to

• Four single-agent trajectories. Each contains 1500 RGB-D frames.
• Four two-agent trajectories. Each is divided into 2 portions, holding 2500 frames, with the exception of Office-0 which includes 1950 frames per part.
• Two pre-trained NetVLAD models for the loop detection module.

Installation

• Method 1 step-by-step set up(Recommended)

• Method 2 Configure the environment in one line

Usage

Run

Evaluation

Acknowledgement

Citation

@misc{hu2023cpslam,
      title={CP-SLAM: Collaborative Neural Point-based SLAM System}, 
      author={Jiarui Hu and Mao Mao and Hujun Bao and Guofeng Zhang and Zhaopeng Cui},
      year={2023},
      eprint={2311.08013},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}