MedBN-robust-test-time-adpatation

[CVPR 2024] Official implementation of MedBN: Robust Test Time Adaptation against Malicious Test Samples by Hyejin Park*, Jeongyeon Hwang*, Sunung Mun, Sangdon Park, and Jungseul Ok

You can visit our project page 🚀!

Requirements

• RobustBench

Datasets

You need to download the CIFAR10-C and CIFAR100-C data to ../data/

Pretrained Model

We utilize a pre-trained model from the DIA repository for the CIFAR dataset, and from RobustBench for the ImageNet dataset.

Usage

Examples for running code for CIFAR10-C

python test_attack.py --cfg ./cfgs/cifar10/[method].yaml \
MODEL.ARCH resnet26 MODEL.NORM [bn/med] ATTACK.STEPS 100 ATTACK.SOURCE 40 

If you have any questions, please contact Hyejin Park by [[email protected]]

Citation

If our MedBN method are helpful in your research, please consider citing our paper:

@inproceedings{park2024medbn,
  title={MedBN: Robust Test-Time Adaptation against Malicious Test Samples},
  author={Park, Hyejin and Hwang, Jeongyeon and Mun, Sunung and Park, Sangdon and Ok, Jungseul},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={5997--6007},
  year={2024}
}

Acknowledgement

The code is inspired by DIA, TENT, EATA, SAR, and SoTTA.