The extension version (B-MTARD) of this paper is accepted by TPAMI, which can be found in Here )

ECCV2022: Enhanced Accuracy and Robustness via Multi-Teacher Adversarial Distillation

The Offical Code of ECCV2022: Enhanced Accuracy and Robustness via Multi-Teacher Adversarial Distillation

by Shiji Zhao, Jie Yu, Zhenlong Sun, Bo Zhang, Xingxing Wei.

For the teacher model, the WideResNet-34-10 TRADES-pretrained and WideResNet-70-16 model follows the setting in RSLAD. The clean teacher model checkpoint is here.

Our checkpoint is here.

the running environment

python=3.8 
pytorch=1.6
cuda = 11.3
numpy=1.19

training resnet18 on cifar10:

python mtard_resnet18_cifar10.py

training resnet18 on cifar100:

python mtard_resnet18_cifar100.py

Citation

@inproceedings{Zhao2022Enhanced,
title={Enhanced Accuracy and Robustness via Multi-Teacher Adversarial Distillation},
author={Shiji Zhao and Jie Yu and Zhenlong Sun and Bo Zhang and Xingxing Wei},
booktitle={European Conference on Computer Vision},
year={2022},
}