README

Building the MVTec Caption Dataset

This document describes how to build the MVTec Caption Dataset. The MVTec Caption Dataset is constructed in two parts, requiring the separate construction of the MVTEC-AD-Caption and MVTEC-LOCO-Caption datasets.

（1）MVTEC-AD-Caption

1. Download the MVTEC-AD dataset from https://www.mvtec.com/company/research/datasets/mvtec-ad
2. Run tools/Construct_MVTEC-AD-Caption.py to build the MVTEC-AD-Caption part of the dataset.

python tools/Construct_MVTEC-AD-Caption.py

（2）MVTEC-LOCO-Caption

1. Download the MVTec LOCO AD dataset from https://www.mvtec.com/company/research/datasets/mvtec-loco
2. Run tools/Construct_MVTEC-LOCO-Caption.py to build the MVTEC-LOCO-Caption part of the dataset.

python tools/Construct_MVTEC-LOCO-Caption.py

Dataset Structure

Expected Structure for MVTEC-AD-Caption

datasets/
	mvtec_anomaly_detection/
	  bottle/
	    ground_truth/
	    prompt/
	    test/
	    train/
	  cable/
	    ground_truth/
	    prompt/
	    test/
	    train/
	  ...
	  zipper/
	    ground_truth/
	    prompt/
	    test/
	    train/

Expected Structure for MVTEC-LOCO-Caption

datasets/
	mvtec_loco_anomaly_detection/
	  breakfast_box/
	    ground_truth/
	    ground_truth_merge_mask/
 	    prompt/
            test/
	    train/
	  juice_bottle/
	    ground_truth/
	    ground_truth_merge_mask/
 	    prompt/
            test/
	    train/
	  ...
	  splicing_connectors/
	    ground_truth/
	    ground_truth_merge_mask/
 	    prompt/
            test/
	    train/

Citation

@article{hu2024anomalyxfusion,
  title={AnomalyXFusion: Multi-modal Anomaly Synthesis with Diffusion},
  author={Hu, Jie and Huang, Yawen and Lu, Yilin and Xie, Guoyang and Jiang, Guannan and Zheng, Yefeng},
  journal={arXiv preprint arXiv:2404.19444},
  year={2024}
}