IG-SLAM Instant Gaussian SLAM (Presented at ECCV'24 NeuSLAM Workshop)

This repository contains the code and trained models of our work "IG-SLAM: Instant Gaussian SLAM", ECCV'24 NeuSLAM

by Furkan Aykut Sarıkamış, A. Aydın Alatan

Department of Electrical and Electronics Engineering, Middle East Technical University

[Paper & Supplementary](https://arxiv.org/abs/2408.01126/)

Abstract

3D Gaussian Splatting has recently shown promising results as an alternative scene representation in SLAM systems to neural implicit representations. However, current methods either lack dense depth maps to supervise the mapping process or detailed training designs that consider the scale of the environment. To address these drawbacks, we present IG-SLAM, a dense RGB-only SLAM system that employs robust dense SLAM methods for tracking and combines them with Gaussian Splatting. A 3D map of the environment is constructed using accurate pose and dense depth provided by tracking. Additionally, we utilize depth uncertainty in map optimization to improve 3D reconstruction. Our decay strategy in map optimization enhances convergence and allows the system to run at 10 fps in a single process. We demonstrate competitive performance with state-of-the-art RGB-only SLAM systems while achieving faster operation speeds. We present our experiments on the Replica, TUM-RGBD, ScanNet, and EuRoC datasets. The system achieves photo-realistic 3D reconstruction in large-scale sequences, particularly in the EuRoC dataset.

Contributions:

• We present IG-SLAM, an efficient dense RGB SLAM system that performs at high frame rates, offering scalability and robustness even in challenging conditions.

• A novel 3D reconstruction algorithm that accounts for depth uncertainty, making the 3D reconstruction robust to noise.

• A training procedure to make dense depth supervision for the mapping process as efficient as possible.

System Overview

IG-SLAM consists of two main parts: tracking and mapping.

Please cite as:

@article{sarikamis2024ig,
  title={Ig-slam: Instant gaussian slam},
  author={Sarikamis, F Aykut and Alatan, A Aydin},
  journal={arXiv preprint arXiv:2408.01126},
  year={2024}
}

Code

You may need to install GLM library.

sudo apt-get install libglm-dev

You can create an anaconda environment called igslam.

git clone --recursive https://github.com/Liouvi/IG-SLAM.git

conda env create -f environment.yml
conda activate igslam
python setup.py install

You may need to close visualization to improve reconstruction results

For example:

configs/Replica/replica_mono.yaml

inherit_from: configs/Replica/base_config.yaml

verbose: False
dataset: 'replica'
mode: mono
stride: 1
only_tracking: False
vis: False

Set vis variable as intended.

Replica

bash scripts/download_replica.sh

python run.py --config configs/Replica/office0_mono.yaml

TUM

bash scripts/download_tum.sh

python run.py --config configs/tum/fr1_desk.yaml

ScanNet

Please follow the data downloading procedure on ScanNet website, and extract color/depth frames from the .sens file using this code.

  datasets
  └── ScanNet
      └── scans
          └── scene0000_00
              └── frames
                  ├── color
                  │   ├── 0.jpg
                  │   ├── 1.jpg
                  │   ├── ...
                  │   └── ...
                  ├── depth
                  │   ├── 0.png
                  │   ├── 1.png
                  │   ├── ...
                  │   └── ...
                  ├── intrinsic
                  └── pose
                      ├── 0.txt
                      ├── 1.txt
                      ├── ...
                      └── ...

Once the data is set up properly, you can run:

python run.py --config configs/ScanNet/scene0000_mono.yaml

EuRoC

bash scripts/download_euroc.sh

The GT trajectory can be downloaded from GO-SLAM's Drive Google Drive.

  DATAROOT
  └── EuRoC
     └── MH_01_easy
         └── mav0
             ├── cam0
             ├── cam1
             ├── imu0
             ├── leica0
             ├── state_groundtruth_estimate0
             └── body.yaml
         └── MH_01_easy.txt

Then you can run:

python run.py --config configs/EuRoC/mh_01_easy.yaml

Qualitative Results

Contacts

For questions, please send an email to aykut.sarikamis@metu.edu.tr

Acknowledgements

This project would not be possible without excellent works including GO-SLAM, NeRF-SLAM, MonoGS, and DROID-SLAM.