SNN_Calibration

Pytorch Implementation of Spiking Neural Networks Calibration, ICML 2021

Paper Version 1: [PMLR], [arXiv].

Paper Version 2: [IJCV].

When converting ANNs to SNNs, conventional methods ensure the minimization in parameter space, while we focus on the minimization in network output space:

introduction_figure

Feature Comparison of SNN calibration:

Features	SNN Direct Training	ANN-SNN Conversion	SNN Calibration
Accuracy (T<100)	High	Low	High
Scalability to ImageNet	Tiny	Large	Large
Training Speed	Slow	Fast	Fast
# Required Data	Full-set (1.2M For ImageNet)	~1000	~1000
Inference Speed	Fast	Slow	Fast

Requirements

Pytorch 1.8

For ImageNet experiments, please be sure that you can initialize distributed environments

For CIFAR experiments, one GPU would suffice.

Update (May 31, 2022): New version of the code

We released a new version of the paper, and will update the code to match the experiments with that paper.

Update (Jan 14, 2022): ImageNet experiments

For imagenet experiments, please first download the checkpoints from Google Drive.

We recommend initializing distributed learning environments, and utlizing multi-GPU calibration.

For reproducibility, 8 GPUs are per run and distributed environments are highly encouraged.

For example:

python main_cal_imagenet.py --arch res34 --T 32 --usebn --calib light --dpath PATH/TO/DATA

Pre-training ANN on CIFAR10&100

Train an ANN model with main_train_cifar.py

python main_train_cifar.py --dataset CIFAR10 --arch VGG16 --usebn

Pre-trained results:

Dataset	Model	Random Seed	Accuracy
CIFAR10	VGG16	1000	95.76
CIFAR10	ResNet-20	1000	95.68
CIFAR100	VGG16	1000	77.98
CIFAR100	ResNet-20	1000	76.52

SNN Calibration on CIFAR10&100

Calibrate an SNN with main_cal_cifar.py.

python main_cal_cifar.py --dataset CIFAR10 --arch VGG16 --T 16 --usebn --calib advanced --dpath PATH/TO/DATA

--T is the time step, --calib is the calibration method, please use none, light, advanced for experiments.

The calibration will run 5 times, and return the mean accuracy as well as the standard deviation.

Example results:

Architecture	Dataset	T	Random Seed	Calibration	Mean Acc	Std.
VGG16	CIFAR10	16	1000	None	64.52	4.12
VGG16	CIFAR10	16	1000	Light	93.30	0.08
VGG16	CIFAR10	16	1000	Advanced	93.65	0.25
ResNet-20	CIFAR10	16	1000	None	67.88	3.63
ResNet-20	CIFAR10	16	1000	Light	93.89	0.20
ResNet-20	CIFAR10	16	1000	Advanced	94.33	0.12
VGG16	CIFAR100	16	1000	None	2.69	0.76
VGG16	CIFAR100	16	1000	Light	65.26	0.99
VGG16	CIFAR100	16	1000	Advanced	70.91	0.65
ResNet-20	CIFAR100	16	1000	None	39.27	2.85
ResNet-20	CIFAR100	16	1000	Light	73.89	0.15
ResNet-20	CIFAR100	16	1000	Advanced	74.48	0.16

If you feel this repo helps you, please consider citing the following articles:

@article{li2022converting,
  title={Converting Artificial Neural Networks to Spiking Neural Networks via Parameter Calibration},
  author={Li, Yuhang and Deng, Shikuang and Dong, Xin and Gu, Shi},
  journal={arXiv preprint arXiv:2205.10121},
  year={2022}
}

@article{li2024error,
  title={Error-Aware Conversion from ANN to SNN via Post-training Parameter Calibration},
  author={Li, Yuhang and Deng, Shikuang and Dong, Xin and Gu, Shi},
  journal={International Journal of Computer Vision},
  pages={1--24},
  year={2024},
  publisher={Springer}
}

yhhhli / SNN_Calibration

readme