TJXTT / SNN2ANN

SNN2ANN: A Fast and Memory-Efficient Training Framework for Spiking Neural Networks

SNN2ANN: A Fast and Memory-Efficient Training Framework for Spiking Neural Networks

The demo code of the paper: SNN2ANN: A Fast and Memory-Efficient Training Framework for Spiking Neural Networks

SNN2ANN training with ReSU:

VGG-13

python run.py --arch=VGG --time-step=4 --batch-size=512 --spike-unit=ReSU --dataset=CIFAR100 --class-nums=100 --data-path=dataset_path --pretrained=pretrained_model_path

ResNet-17

python run.py --arch=ResNet --time-step=5 --batch-size=512 --spike-unit=ReSU --dataset=CIFAR100 --class-nums=100 --kaiming-norm=True --data-path=dataset_path --pretrained=pretrained_model_path

SNN2ANN training with STSU:

VGG-13

python run.py --arch=VGG --time-step=4 --batch-size=512 --spike-unit=STSU --dataset=CIFAR100 --class-nums=100 --data-path=dataset_path --pretrained=pretrained_model_path

ResNet-17

python run.py --arch=ResNet --time-step=5 --batch-size=512 --spike-unit=STSU --dataset=CIFAR100 --class-nums=100 --kaiming-norm=True --data-path=dataset_path --pretrained=pretrained_model_path

Pre-trained Model

The pre-trained models are available at https://drive.google.com/drive/folders/1y6ZUT3WToowuCo72CVspurQrdSG4U8zi?usp=sharing

Citation

@article{snn2ann,
  title={SNN2ANN: A Fast and Memory-Efficient Training Framework for Spiking Neural Networks},
  author={Jianxiong Tang and Jianhuang Lai and Xiaohua Xie and Lingxiao Yang and Wei-Shi Zheng},
  journal={arXiv preprint arXiv:2206.09449},
  year={2022}
}