STDL

This repo contains the PyTorch implementation of the paper: Spatio-Temporal Decoupled Learning for Spiking Neural Networks.

Usage

Dependencies

• pytorch

Scripts

• Train ResNet18 with STDL on CIFAR10:

python main_train.py --lr 0.4 --batch_size 512 --cos_lr --epochs 400 --pos_bn 3 --arch spk_cifar_resnet_local --net spk_resnet18 --wd 5e-5 --decay 0.1 --threshold 1.0 --time_window 4 --aux_net_config mem --detach_mem --detach_reset --cutout --auto_aug --local_module_num 3 --rule mem_1800 

• Train ResNet19 with STDL on CIFAR100:

python main_train.py --dataset cifar100 --lr 0.4 --batch_size 512 --cos_lr --epochs 400 --pos_bn 3 --arch spk_cifar_resnet_local --net spk_resnet19 --wd 5e-5 --decay 0.1 --threshold 1.0 --time_window 4 --aux_net_config mem --detach_mem --detach_reset --cutout --auto_aug --local_module_num 3 --rule mem_4400 

Please refer to run.sh for the scripts of more datasets (ImageNet and CIFAR10DVS) and baselines (ELL and DECOLLE).

Contact

If you have any questions, please contact chenxiang.ma@connect.polyu.hk.