Asynchronous Convolutional Networks for Object Detection in Neuromorphic Cameras

Code for the CVPR2019 Event-based Vision Workshop paper "A Differentiable Recurrent Surface for Asynchronous Event-Based Data"
Authors: Marco Cannici, Marco Ciccone, Andrea Romanoni, Matteo Matteucci

Citing:

If you use this work for research, please cite our accompanying CVPR2019 Event-based Vision Workshop paper:

@inproceedings{cannici2019asynchronous,
  title={Asynchronous Convolutional Networks for Object Detection in Neuromorphic Cameras},
  author={Cannici, Marco and Ciccone, Marco and Romanoni, Andrea and Matteucci, Matteo},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year={2019}
}

Requirements

• TensorFlow 1.4.0
• Cython extensions: build_setup.sh

You can create a conda environment to run the code as it follows:

conda create -n aync-ev-cnn python=3.6`
conda activate aync-ev-cnn
conda env update -f=requirements.yml
python cython_setup.py build_ext --inplace

Run scripts

• To check event layers equivalence (no dataset or checkpoint required):
  python src/scripts/test_correctness.py

• To run network predictions on a dataset (select the proper .yml file):
  

  • Unzip under data/ the N-Caltech101 data and checkpoint
  • python src/scripts/run_networks.py -c configs/efcn_event.yml