This research introduces ALFA, a novel approach that enhances large vision-language models' ability to detect visual anomalies without prior training on anomalous examples.

• Overcomes the limitations of static prompts by using adaptive prompting
• Focuses on local feature analysis rather than just global image assessment
• Achieves superior performance in zero-shot anomaly detection and localization
• Demonstrates practical applications across industrial inspection and security monitoring

For security applications, this advancement enables more reliable surveillance systems that can identify threats or unusual activities with greater precision and fewer false positives, even when facing previously unseen anomaly types.

Do LLMs Understand Visual Anomalies? Uncovering LLM's Capabilities in Zero-shot Anomaly Detection