Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Towards Efficient Pixel Labeling for Industrial Anomaly Detection and Localization

Created by
  • Haebom

Author

Jingqi Wu, Hanxi Li, Lin Yuanbo Wu, Hao Chen, Deyin Liu, Peng Wang

Outline

This paper proposes ADClick, an interactive image segmentation (IIS) algorithm for industrial product inspection. ADClick significantly improves the performance of anomaly detection models by generating pixel-level anomaly detection annotations with just a few user clicks and brief text descriptions, without pixel-level annotations of defective samples (e.g., AP = 96.1% on MVTec AD). Furthermore, we introduce ADClick-Seg, a multimodal framework that aligns visual features and text prompts using a prototype-based approach. By combining pixel-level prior information with linguistic guidance cues, ADClick-Seg achieves state-of-the-art results on the challenging "multi-class" anomaly detection task (AP = 80.0%, PRO = 97.5%, Pixel-AUROC = 99.1% on MVTec AD).

Takeaways, Limitations

•
Takeaways:
◦
Generate efficient and accurate anomaly detection annotations without pixel-level annotations.
◦
Precise anomaly detection possible with just user clicks and text descriptions.
◦
Effectively leverage visual features and text prompts through a multi-modal framework.
◦
Achieving state-of-the-art performance on the MVTec AD dataset
•
Limitations:
◦
Additional experiments are needed to evaluate the generalization performance of the proposed method.
◦
Need to evaluate applicability to various industrial environments and product types
◦
Need to analyze performance changes according to the number of user clicks or the quality of text descriptions
👍
Made with Slashpage