Search papers, labs, and topics across Lattice.
This paper introduces a Dual-Anchor framework for zero-shot anomaly detection (ZSAD) that enhances performance by integrating hierarchical image anchors alongside traditional text anchors. By employing a top-down grouping mechanism to aggregate image features, the method creates robust normal and abnormal group tokens that improve visual grounding and reduce sensitivity to prompt design. The framework demonstrates significant improvements in image-text alignment and generalization across multiple industrial and medical benchmarks, outperforming existing CLIP-based approaches.
Hierarchical image anchors can dramatically enhance zero-shot anomaly detection, reducing prompt sensitivity and improving alignment across diverse domains.
Zero-shot anomaly detection (ZSAD) aims to identify anomalies in unseen domains, a setting that is particularly critical for industrial and medical applications where domain shifts are prevalent. However, most CLIP-based ZSAD methods anchor semantics solely on the text modality, making performance highly sensitive to prompt design and leading to weak visual grounding. To mitigate these limitations, we propose a Dual-Anchor framework that complements conventional text anchors with hierarchical image anchors constructed via a top-down grouping mechanism. This mechanism progressively aggregates local-to-global image features to form normal and abnormal group tokens, which serve as image anchors and act as gating signals in a Group-Gated Token Refiner to enhance the global representation. The refined image anchors are then fused with text prompts to construct dynamic state prompts. By jointly reinforcing visual and textual semantics, our framework stabilizes image-text alignment, reduces prompt dependency, and achieves strong generalization across 8 industrial and 6 medical benchmarks.
