Core Concepts
Utilizing CLIP-ADA for anomaly detection enhances localization accuracy and achieves state-of-the-art results.
Abstract
The paper introduces CLIP-ADA, a framework for anomaly detection using pre-trained CLIP models. It focuses on unified anomaly detection across multiple categories by introducing learnable prompts and a region refinement strategy. The framework outperforms existing methods in anomaly detection and localization tasks, showcasing its effectiveness in adapting CLIP to industrial images.
Stats
Achieved state-of-the-art 97.5/55.6 and 89.3/33.1 on MVTec-AD and VisA for anomaly detection and localization.
Extensive experiments demonstrate the superiority of the framework.
Quotes
"We propose a simple yet effective approach to get a unified representation across diverse image categories."
"Our method identifies anomaly regions more faithfully than compared methods."