SpikingEDN

This code is a demo of our TNNLS 2024 paper "Accurate and Efficient Event-based Semantic Segmentation Using Adaptive Spiking Encoder-Decoder Network".

Dataset

To proceed, please download the DDD17/DSEC-SEMANTIC dataset on your own.

Environment

1. Python 3.8.*
2. CUDA 10.0
3. PyTorch 
4. TorchVision 
5. fitlog

Install

Create a virtual environment and activate it.

conda create -n SpikingEDN python=3.8
conda activate SpikingEDN

The code has been tested with PyTorch 1.6 and Cuda 10.2.

conda install pytorch=1.6.0 torchvision=0.7.0 cudatoolkit=10.2 -c pytorch
conda install matplotlib path.py tqdm
conda install tensorboard tensorboardX
conda install scipy scikit-image opencv

Code for SpikingEDN

We provide retrain and evaluate code for DDD17/DSEC-SEMANTIC. The best model is provided on '/logs/retrain/retrain_best_model/encoder_best_model'

Retrain

For retrain procedure, execute:
bash retrain_ddd17.sh

Evaluate

For evaluate procedure, execute:
bash evaluate_best_model_ddd17.sh bash evaluate_dsec.sh

Paper Reference

@article{zhang2024accurate,
  title={Accurate and efficient event-based semantic segmentation using adaptive spiking encoder--decoder network},
  author={Zhang, Rui and Leng, Luziwei and Che, Kaiwei and Zhang, Hu and Cheng, Jie and Guo, Qinghai and Liao, Jianxing and Cheng, Ran},
  journal={IEEE Transactions on Neural Networks and Learning Systems},
  year={2024},
  publisher={IEEE}
}

Our code is developed based on the code from papers "Differentiable hierarchical and surrogate gradient search for spiking neural networks" and "Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation"

code:
https://github.com/Huawei-BIC/SpikeDHS https://github.com/NoamRosenberg/autodeeplab

License

This open-source project is not an official Huawei product, and Huawei is not expected to provide support for this project.