Researchers have developed PromptAD, a novel method for few-shot anomaly detection, which traditionally requires crafting numerous prompts through prompt engineering for industrial scenarios. This innovative approach utilizes semantic concatenation to transform normal prompts into anomaly prompts by appending specific suffixes, creating numerous negative samples to facilitate prompt learning in a one-class setting.

Furthermore, to overcome the hurdle of training without actual anomaly images, PromptAD introduces an explicit anomaly margin. This hyper-parameter-driven mechanism precisely regulates the distinction between normal and anomaly prompt features. In image-level and pixel-level anomaly detection, PromptAD has achieved top results in 11 out of 12 few-shot settings on MVTec and VisA datasets.

• Highlights from the paper include:
  • Introduction of a one-class prompt learning method
  • Proposal of semantic concatenation for prompt transformation
  • Implementation of explicit anomaly margin for optimal feature separation
  • Record-setting performance in few-shot anomaly detection scenarios

This work represents an essential step forward in AI’s role in industrial applications, specifically in anomaly detection. It is indicative that fewer and more generically applicable prompts can achieve excellent results, streamlining the anomaly detection process and making it more accessible. The PromptAD approach opens pathways for future one-class learning scenarios, potentially revolutionizing the field of computer vision anomaly detection.