This repository contains code for our paper "How Many Events Do You Need? Event-based Visual Place Recognition Using Sparse But Varying Pixels".

If you use this code, please refer to our paper:

@article{FischerRAL2022ICRA2023,
    title={How Many Events do You Need? Event-based Visual Place Recognition Using Sparse But Varying Pixels},
    author={Tobias Fischer and Michael Milford},
    journal={IEEE Robotics and Automation Letters},
    volume={7},
    number={4},
    pages={12275--12282},
    year={2022},
    doi={10.1109/LRA.2022.3216226},
}

The associated QCR-Event-VPR dataset can be found on Zenodo. The code can also handle data from our previous Brisbane-Event-VPR dataset.

Please download the dataset, and place the parquet files into the ./data/input_parquet_files directory. If you want to work with the DAVIS conventional frames, please download the zip files, and extract them so that an image files is located in e.g. ./data/input_frames/bags_2021-08-19-08-25-42/frames/1629325542.939281225204.png. For the Brisbane-Event-VPR dataset, place the nmea files into the ./data/gps_data directory.

We recommend using conda (in particular, mamba, which can be installed via Miniforge3:

mamba create -n sparse-event-vpr pip pytorch codetiming tqdm pandas numpy scipy matplotlib seaborn numba pynmea2 opencv pypng h5py importrosbag pbr pyarrow fastparquet
mamba activate sparse-event-vpr
pip install git+https://github.com/Tobias-Fischer/tonic.git@develop --no-deps

First, install the package via

pip install -e .  # you need to run this command inside the `sparse-event-vpr` directory

The main script file is perform_sparse_event_vpr. You can run it with Python and see all options that are exposed:

python ./scripts/perform_sparse_event_vpr.py --help

Furthermore, there is a standalone Jupyter notebook available that guides you through some of the key concepts of the paper.