Problem
Research questions and friction points this paper is trying to address.
zero-shot anomaly detection
CLIP
prompt dependency
visual grounding
domain shift
Dual Anchors, Do It Better: Hierarchical Group Merging for Zero-Shot Anomaly Detection
🤖 AI Summary
This work addresses the instability of semantic alignment, weak visual representation, and high sensitivity to textual prompts in zero-shot anomaly detection. To mitigate these issues, the authors propose a dual-anchor framework that complements the conventional text anchor with a novel image anchor derived from a top-down hierarchical image grouping strategy, serving as a visual guidance signal. A group-gated token refiner is introduced to enhance global representation, while dynamic state prompting is integrated to jointly strengthen vision-language semantic alignment. Built upon the CLIP architecture, the proposed method substantially reduces reliance on textual prompts and achieves state-of-the-art performance across eight industrial and six medical benchmark datasets, demonstrating significantly improved cross-domain generalization and robustness.
📝 Abstract
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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Zero-shot anomaly detection
Dual-anchor framework
Hierarchical group merging
Image-text alignment
Prompt engineering
🔎 Similar Papers
No similar papers found.
