FSR: A feature shuffling and restoration strategy for universal unsupervised anomaly detection (KBS 2025)

PyTorch implementation and for KBS2025 paper, A feature shuffling and restoration strategy for universal unsupervised anomaly detection.

Download Datasets

Please download MVTecAD dataset from MVTecAD dataset and BTAD dataset from BTAD dataset.

Installation

timm==0.3.2
pytoch==1.8.1

Citation

If you find this repository useful, please consider citing our work:

@article{luo2025feature,
  title={A feature shuffling and restoration strategy for universal unsupervised anomaly detection},
  author={Luo, Wei and Yao, Haiming and Qiang, Zhenfeng and Zhang, Xiaotian and Zhang, Weihang},
  journal={Knowledge-Based Systems},
  pages={114874},
  year={2025},
  publisher={Elsevier}
}