Unlocking the Potential of Reverse Distillation for Anomaly Detection

[AAAI 2025] PyTorch Implementation of "Unlocking the Potential of Reverse Distillation for Anomaly Detection". paper

1. Environment

Create a new conda environment firstly.

conda create -n newRD python=3.8
conda activate newRD
pip install -r requirements.txt

2. Prepare Data

MVTec AD Dataset

Download MVTec AD from MVTec AD. Unzip the file to ./data/.

|--data
    |-- mvtec_anomaly_detection
        |-- bottle
        |-- cable
        |-- ....

Describable Textures Dataset

Refer to DRAEM, download Describable Textures dataset from Describable Textures dataset for anomaly synthesis. Unzip the file to ./data/.

|--data
    |-- dtd
        |-- images
        |-- ....

3.Train and Test

To get the training and inference results, simply execute the following command.

python train.py

Acknowledgement

Thanks to the codes provided by Reverse Distillation which greatly support our work.

Citation

If you think this work is helpful to you, please consider citing our paper.

@inproceedings{liu2025unlocking,
  title={Unlocking the potential of reverse distillation for anomaly detection},
  author={Liu, Xinyue and Wang, Jianyuan and Leng, Biao and Zhang, Shuo},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={6},
  pages={5640--5648},
  year={2025}
}