Omni-AD: Learning to Reconstruct Global and Local Features for Multi-class Anomaly Detection

🍭🍭🍭 Our Omni-AD has been accepted at the ICME 2025 conference 🍭🍭🍭

Overview

🧭 Instructions 🧭

1. Data Preparation

The following datasets are required for the project:

• MVTEC AD
• VisA
• Real-IAD

For each dataset:

• Place the data_json/[dataset]/meta.json file into the corresponding dataset's root directory.
• For more details, refer to: Additional Information.

2. Prepare Environment

Set up the Conda environment using the requirements.yml file.

3. Configure Dataset Path

Specify the dataset paths by setting self.data.root in the file:
configs/omniad/dataset_configs.py.

4. Training and Inference

Run the following command for training and inference. You can specify additional settings in run.py:

python run.py

👉More Details

Please refer to ADer

🥰 Citation

@misc{quan2025omniadlearningreconstructglobal,
      title={Omni-AD: Learning to Reconstruct Global and Local Features for Multi-class Anomaly Detection}, 
      author={Jiajie Quan and Ao Tong and Yuxuan Cai and Xinwei He and Yulong Wang and Yang Zhou},
      year={2025},
      eprint={2503.21125},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.21125}, 
}

🙏Acknowledgements

We would like to express our gratitude to the outstanding works of MambaAD and ADer, among others, for their significant contributions and support to our project.