EVLoc

KO | EN

gitlite — search

#react #ai #typescript #llm #go #golang #machine-learning #deep-learning #docker #java #windows #cli #security #mcp

(★ 12)

[ICRA 2025] Official code for "EVLoc: Event-based Visual Localization in LiDAR Maps via Event-Depth Registration"

EVLoc Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# EVLoc
This repository contains the source code for our paper:

[EVLoc: Event-based Visual Localization in LiDAR Maps via Event-Depth Registration](https://arxiv.org/pdf/2503.00167) 
Kuangyi Chen, Jun Zhang, Friedrich Fraundorfer

A short introduction is provided in the video below. 
<a href="https://youtu.be/tQw8s4iZgvg">
 <img src="./media/overview.png" width="480" alt="Watch the video">
</a>

## Updates
This work is extended by our follow-up paper:

[[ICRA 2026] LEAR: Learning Edge-Aware Representations for Event-to-LiDAR Localization](https://arxiv.org/pdf/2603.01839) 
Kuangyi Chen, Jun Zhang, Yuxi Hu, Yi Zhou, Friedrich Fraundorfer

Compared with EVLoc, LEAR further improves event-to-LiDAR localization by learning edge-aware representations and introducing cross-task feature fusion. Code: [LEAR](https://github.com/EasonChen99/LEAR)

## Requirements
The code has been trained and tested with PyTorch 2.2.2 and Cuda 12.2.
```Shell
conda create -n EVLoc python=3.10 -y
conda activate EVLoc
pip install -r requirements.txt
cd blender-mathutils
python setup.py install
cd ..
cd core/correlation_package
python setup.py install
cd ..
cd visibility_package
python setup.py install
cd ../..
```
Additionally, please compile [PoseLib](https://github.com/PoseLib/PoseLib) for use with pose solvers.

## Data Preprocessing
To evaluate/train EVLoc, you could download the M3ED dataset.
* [M3ED](https://m3ed.io/data_overview/)

We trained and tested EVLoc on the M3ED Falcon Indoor sequences flight_1, flight_2, and flight_3.

To obtain the required data for direct use, we sample a pre-defined range of point cloud from the whole LiDAR maps for every provided pose.
```Shell
python tools/map2pc.py --dataset [PATH_TO_DATA] --sequence [SPECIFIED_SEQUENCE] --save_dir [PATH_TO_SAVEDIR]
```
As for event data, we also need to generate event frames accordingly.
```Shell
python tools/event2frame.py --dataset [PATH_TO_DATA] --sequence [SPECIFIED_SEQUENCE] --save_dir [PATH_TO_SAVEDIR] --camera [WHICH_CAMERE] --method [EVENT_REPRESENTATION]
```

## Demos
Pretrained models can be downloaded from [google drive](https://drive.google.com/drive/folders/1ASoSb4XsNDopaBkD503m9Vo_UHYpGTsa?usp=sharing)

You can demo a trained model on a sequence of frames
```Shell
python main_event_baseline.py --ev_input ours_denoise_pre_100000 --load_checkpoints checkpoints/checkpoint.pth -e
```

## Training
You can train a model using `main_event_baseline.py`. Training logs will be written to the `runs` which can be visualized using tensorboard.
```Shell
python main_event_baseline.py --data_path [PATH_TO_PREPROCESSED_DATA] --test_sequence falcon_indoor_flight_3 --max_depth 10. --epochs 100 --batch_size 2 --lr 4e-5 --gpus 0 --max_r 5. --max_t 0.5 --evaluate_interval 1
```
If you want to train a model with the offset alleviation module, you can use `main_event_offset_RT.py`.
```Shell
python main_event_offset_RT.py --data_path [PATH_TO_PREPROCESSED_DATA] --test_sequence falcon_indoor_flight_3 --max_depth 10. --epochs 100 --batch_size 2 --lr 4e-5 --gpus 0 --max_r 5. --max_t 0.5 --evaluate_interval 1
```

EVLoc

EVLoc Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

EVLoc

EVLoc Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README