EFH

The source code for paper: Event-Fused Hybrid ANN-SNN Architecture for Low-Latency Object Detection in Automotive Vision

Method

Prepare python env

python=3.8.13

# install torch
pip install torch==2.0.0+cu117 torchvision==0.15.0+cu117 torchaudio==2.0.0+cu117 --extra-index-url https://download.pytorch.org/whl/cu117
# install yolox
bash libs/download_install.sh
# install packages
pip install -r requirements.txt

Prepare datasets

The DSEC Dataset can download from DSEC-detection, follow DAGr to prepare the dataset

Expected dataset structure for DSEC-detection :

dsec/
    ├── {train,test}/
        ├── interlaken_00_c/
            ├── events/left/
                └── events_2x.h5
            ├── images/left/distorted/
                └── *.png
    ├── train_val_test_spilt.xml

PKU-DAVIS-SOD

The PKU-DAVIS-SOD Dataset can download from PKU-DAVIS-SOD，follow

python ./data/davis346_temporal_event_to_npy.py
python ./data/davis346_to_images.py

convert the raw .aedat4 files to synrhronous frames (.png), events (.npy) and the pairs dir.

Expected dataset structure for PKU_DAVIS_SOD:

 PKU_DAVIS_SOD/
    ├── aps_frames/
        ├── {train,val,test}/
            ├── {normal,low_light,motion_blur}/
                └── 001_train_normal/
                    └── *.png
    └── events_npys/
        ├── {train,val,test}/
    └── annotations/
        ├── {train,val,test}/
    └── pairs/
        ├── {train,val,test}/

Train

train_RsHN.py

Eval

Partial model weights can be downloaded from Google Drive.

eval_on_dsec.py
eval_on_pku.py

Citation

If our work help to your research, please cite our paper, thx.

@ARTICLE{11386767,
  author={Zhang, Chengjun and Zhang, Yuhao and Yu, Jisong and Yang, Jie and Sawan, Mohamad},
  journal={IEEE Robotics and Automation Letters}, 
  title={Event-Fused Hybrid ANN-SNN Architecture for Low-Latency Object Detection in Automotive Vision}, 
  year={2026},
  volume={11},
  number={3},
  pages={3622-3628},
  keywords={Object detection;Cameras;Feature extraction;Low latency communication;Vehicle dynamics;Robot vision systems;Event detection;Autonomous vehicles;Training;Standards;Computer vision for automation;deep learning for visual perception;deep learning methods},
  doi={10.1109/LRA.2026.3662637}}

Thanks to these amazing projects:

• dagr
• dsec-det
• spikingjelly
• YOLOX