SEED: Sparse Convolutional Recurrent Learning for Efficient Event-based Neuromorphic Object Detection

Citation

This is the official implementation for SEED. Please refer to and cite the following paper:

@inproceedings{seed2025,
  title={Sparse Convolutional Recurrent Learning for Efficient Event-based Neuromorphic Object Detection},
  author = {Wang, Shenqi and Xu, Yingfu and Yousefzadeh, Amirreza and Eissa, Sherif and Corporaal, Henk and Corradi, Federico and Tang, Guangzhi},
  booktitle={2025 International Joint Conference on Neural Networks (IJCNN)},
  pages={1--8},
  year={2025},
  organization={IEEE}
}

Dependencies

Prophesee Metavision SDK

We are using Metavision 4.2.1.

Install Prophesee Metavision SDK using this LINK

Prophesee 1 Megapixel Event Dataset

Download DAT dataset from HERE.

Follow the tutorial HERE to pre-process the DAT file to HDF5 file, the pre-processing method used in this project is multi_channel_timesurface.

The pre-comupted dataset can be downloaded HERE. However, we noticed that the content of DAT and pre-computed are different. Our results are trained and tested on the preprocessed DAT dataset.

Enviornment setup

python -m pip install -r requirements.txt

Training and Testing

Training

Please modify the dataset_path in the train.py file to your own dataset path.

Use command python train.py to run the training script.

Validation

Please modify the saved_model_path in the validate.py file to saved model directory.

Please modify the dataset_path in the validate.py file to your own dataset path.

Use command python validate.py to run the training script.

Test

Please modify the saved_model_path in the test.py file to saved model directory.

Select the best performance mode drom the validation and modify the test_epoch in test.py.

Please modify the dataset_path in the test.py file to your own dataset path.

Use command python test.py to run the training script.