💎 EventGeM: Global-to-Local Feature Matching for Event-Based Visual Place Recognition

This repository contains the code for Event-GeM — an event-based visual place recognition (VPR) pipeline that uses a pre-trained transformer vision backbone for feature extraction, generalized mean pooling, and keypoint detection for 2D homology match re-ranking.

Event-GeM uses features from the Event-Camera-Data-Pre-Training and a generalized mean (GeM) pooling layer to generate initial matches from event frames. SuperEvent then allows for 2D homology re-ranking of the TopK matches based on keypoint selection for improved recall performance. Optional additional re-ranking can be performed with Depth AnyEvent using a structural similarity index. Datasets for VPR are managed and generated using Event-LAB.

Getting Started 🚀

Event-GeM is powered by Pixi for all dependency and package management. If not already installed, run the following in your command terminal:

curl -fsSL https://pixi.sh/install.sh | sh

For more information, please see the pixi documentation.

Next, clone our repository with all the required submodules and navigate to the project directory by running the following in your command terminal:

git clone git@github.com:AdamDHines/Event-GeM.git eventgem --recurse-submodules && cd eventgem

Once installed, you can quickly try Event-GeM with our demo by running the following in your command terminal:

pixi run demo

Running EventGeM & EventGeM-D

Basic operation

To run EventGeM and EventGeM-D, you simply need to input a dataset, reference, and query that you would like to run in a single command-line invocation:

pixi run eventgem --dataset brisbane_event --reference sunset2 --query sunset1

This will generate all of the event-based frame types needed for the various backbones. If you wish to save on disk space, you can instead stream from the event file:

pixi run eventgem --dataset brisbane_event --reference sunset2 --query sunset1 --stream

Event-LAB handles the generation of a pseudo-ground truth file for Recall@K evaluation, which runs automatically.

By default, the EventGeM method will run, however if you want to run the EventGeM-D depth based re-rank run:

pixi run eventgem --dataset brisbane_event --reference sunset2 --query sunset1 --method eventgem-d

List of arguments

Dataset parameters

• --dt-ms: max time in msec to consturct polarity and tencode frames to (default=50)
• --mcts-time: list of times to construct the MCTS frames with (deafult=[10, 20, 30, 40, 50])
• --time-scale: time scale of the event timestamps (default=1e-9)
• --start-time: delay time for streaming from event files for synchronizing (default=None)
• --skip: number of frames to skip during inference (default=None)

Model parameters

• --stream: stream from event file instead of generating and saving event frames
• --top-k: top-k reference to run the re-ranking on (default=50)
• --se-topk: number of keypoints to detect per frame and run RANSAC on (default=170)
• --backbone-batch-size: batch size for running the initial ViT embeddings (non-stream mode only)
• --keypoint-batch-size: batch size for running the keypoint detection (non-stream mode only)
• --rerank-mode: run keypoint, depth, or both re-rank options (default="keypoints")
• --method: VPR method to run during inference (default="eventgem")

Directory parameters

• --data-root: default root directory to store datasets (default="./eventgem/data")
• --feature-out: default directory for ViT embedding features (default="./eventgem/features")
• --keypoint-out: default directory for keypoints (default="./eventgem/keypoints")
• --depth-out: default directory for depth maps (default="./eventgem/depth")

Citation

If you found our work interesting or use it as a baseline method, please cite the following:

@misc{hines2026eventgem,
      title={EventGeM: Global-to-Local Feature Matching for Event-Based Visual Place Recognition}, 
      author={Adam D. Hines and Gokul B. Nair and Nicolás Marticorena and Michael Milford and Tobias Fischer},
      year={2026},
      eprint={2603.05807},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.05807}, 
}

Contributing and Issues

If you encounter any issues or want to contribute a fix, please open an issue or a pull request.