arclab-hku/DEIO — Gitpedia

<p align="center"> <h1 align="center"> DEIO: Deep Event Inertial Odometry </h1> <h3 align="center"> <a href="https://arxiv.org/abs/2411.03928">Paper</a> | <a href="https://kwanwaipang.github.io/DEIO">Website</a> | <a href="https://www.youtube.com/watch?v=gs_LLOh3AsQ">Demo</a> </h3>  <div align="center"> <img src="./Figs/cover_figure.png" width="90%" /> </div>

Abstract

<div align="justify"> Event cameras show great potential for visual odometry (VO) in handling challenging situations, such as fast motion and high dynamic range. Despite this promise, the sparse and motion-dependent characteristics of event data continue to limit the performance of feature-based or direct-based data association methods in practical applications. To address these limitations, we propose Deep Event Inertial Odometry (DEIO), the first monocular learning-based event-inertial framework, which combines a learning-based method with traditional nonlinear graph-based optimization. Specifically, an event-based recurrent network is adopted to provide accurate and sparse associations of event patches over time. DEIO further integrates it with the IMU to recover up-to-scale pose and provide robust state estimation. The Hessian information derived from the learned differentiable bundle adjustment (DBA) is utilized to optimize the co-visibility factor graph, which tightly incorporates event patch correspondences and IMU pre-integration within a keyframe-based sliding window. Comprehensive validations demonstrate that DEIO achieves superior performance on 10 challenging public benchmarks compared with more than 20 state-of-the-art methods.  </div>

Update log

README Upload (2024/10/28)
Paper Upload (2024/11/06)
Estimated Trajectories Upload (2024/11/07)
Code Upload (2025/07/19)
More Raw Results of VECtor Dataset (2025/07/20)

Setup and Installation

sh
# for cuda 11.7
conda env create -f environment.yml   
conda activate DEIO
# conda remove --name DEIO --all

pip install .
pip install numpy-quaternion==2022.4.3

# install GTSAM
cd thirdparty/gtsam
mkdir build
cd build
cmake .. -DGTSAM_BUILD_PYTHON=1 -DGTSAM_PYTHON_VERSION=3.10.11
make python-install

Run an Example

sh
conda activate DEIO

# example for the davia240c
CUDA_VISIBLE_DEVICES=0 PYTHONPATH={YOUR_PATH}/DEIO python script/eval_deio/davis240c.py \
    --inputdir=/media/lfl-data2/davis240c \
    --config=config/davis240c.yaml \
    --val_split=script/splits/davis240c/davis240c_val.txt \
    --enable_event \
    --network={YOUR_PATH}/DEVO/DEVO.pth  \
    --plot \
    --save_trajectory \
    --trials=5

The Results will save in path: results.

Using Our Results as Comparison

<div align="justify"> For the convenience of the comparison, we release the estimated trajectories of DEIO in <code>tum</code> format in the dir of <code>estimated_trajectories</code>. What's more, we also give the <a href="./estimated_trajectories/evo_evaluation_trajectory.ipynb">sample code</a> for the quantitative and qualitative evaluation using <a href="https://github.com/MichaelGrupp/evo">evo package</a> </div>

Acknowledgement

This work is based on DPVO, DEVO, DROID-SLAM, DBA-Fusion, and GTSAM
If you find this work helpful in your research, a simple star or citation of our work should be the best affirmation for us. :blush:

@inproceedings{GWPHKU:DEIO,
  title={Deio: Deep event inertial odometry},
  author={Guan, Weipeng and Lin, Fuling and Chen, Peiyu and Lu, Peng},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={4606--4615},
  year={2025}
}