The Industrial Screen Printing Anomaly Detection Dataset (ISP-AD) 机器收录·待核验

mxdsh4pe91xj1754

Research on industrial anomaly detection is limited by the availability of datasets that capture imperfect imaging conditions and complex defect characteristics. To fill this gap, we present the Industrial Screen Printing Anomaly Detection dataset (ISP-AD), which represents a real-world industrial use case in screen printing. It features subtle, weakly contrasted surface defects embedded within structured patterns with high permitted design variability. Comprising a total of 559,049 samples, this dataset is the largest publicly available industrial anomaly detection dataset to date, enabling both unsupervised and supervised training scenarios.

ISP-AD consists of:

312,674 fault-free samples

246,375 defective samples:

245,664 synthetic defects

711 real defects

Designed to advance research in unsupervised, self-supervised, and supervised anomaly detection, ISP-AD provides a benchmark for evaluating defect detection methods under realistic industrial conditions. Additional fault-free data splits enable further investigation of emerging defect synthesis approaches.

This dataset accompanies the paper:

ISP-AD: a large-scale real-world dataset for advancing industrial anomaly detection with synthetic and real defects published in the Journal of Intelligent Manufacturing

Detailed information on dataset generation, specifications, data splits and intended use cases can be found in the accompanying paper. The dataset structure is visualized in the README.md file.

Official GitHub repository (https://github.com/p4ulk/isp-ad) is now available, containing dataloaders based on PyTorch and Anomalib, as well as minimal working examples.

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC-BY-NC-SA 4.0) License. To view a copy of the license, visit https://creativecommons.org/licenses/by-nc-sa/4.0/

Attribution

If you use this dataset in scientific work, please cite the paper as the primary r

本卡描述如含来自官方页的原文片段,其版权归原作者,不在本站 CC-BY 4.0 许可范围(见关于与许可)。

落地页

https://doi.org/10.5281/zenodo.14911042

许可证

CC-BY-NC-SA-4.0 (置信:verified_official)

发布年份

2025

发布方

Zenodo

分发点