Binary Event-Driven Spiking Transformer (IJCAI)

Highlights

• Pure Binary Spiking Computation: All internal activations are binary spike events.
• Fully Event-Driven Attention
• Binary Weight

---

Reference

If you find this repository useful, please consider citing:

@inproceedings{ijcai2025p458,
  title     = {Binary Event-Driven Spiking Transformer},
  author    = {Cao, Honglin and Zhou, Zijian and Wei, Wenjie and Liang, Yu and Belatreche, Ammar and Zhang, Dehao and Zhang, Malu and Yang, Yang and Li, Haizhou},
  booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on
               Artificial Intelligence, {IJCAI-25}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {James Kwok},
  pages     = {4110--4118},
  year      = {2025},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2025/458},
  url       = {https://doi.org/10.24963/ijcai.2025/458},
}

---

Requirements

timm==0.6.12
cupy==11.4.0
torch==1.12.1
spikingjelly==0.0.0.0.12
pyyaml

---

Dataset Preparation

ImageNet-1K

The ImageNet dataset should be organized as follows:

imagenet/
├── train/
│   ├── n01440764/
│   │   ├── n01440764_10026.JPEG
│   │   ├── ...
│   ├── ...
├── val/
│   ├── n01440764/
│   │   ├── ILSVRC2012_val_00000293.JPEG
│   │   ├── ...
│   ├── ...

You may extract ImageNet using this script: https://gist.github.com/BIGBALLON/8a71d225eff18d88e469e6ea9b39cef4

---

Training

Training on ImageNet

Configure hyper-parameters in imagenet.yml:

cd imagenet
./train.sh

Training on CIFAR

Configure hyper-parameters in cifar10.yml or cifar100.yml. You can download the teacher checkpoint from Release model_best.pth-16.tar && model_best.pth-100.tar · CaoHLin/BESTFormer:

cd cifar
./train.sh

---

---

Acknowledgement

This project is built upon and inspired by:

• Spikingformer
• Spikformer
• SpikingJelly