DuoAD: Leveraging [CLS] Dual Characteristics for Training-Free Few-Shot Anomaly Detection
2026-07-27 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors noticed that a special part of Vision Transformers called the [CLS] token can both represent the overall scene and highlight strange areas in images. They used this insight to create an automatic anomaly detection method that needs no training or manual tuning. Their approach picks the best image augmentations and adjusts which parts of the image to focus on based on the attention from the [CLS] token. This method works well on various datasets and is easy to use without extra setup.
Vision TransformersAnomaly Detection[CLS] tokenAttention MapsImage AugmentationFeature ReweightingOne-shot LearningTraining-freeMVTec-AD
Authors
Jyun-Ze Tang, Po-Han Huang, Ming-Ching Chang, Chih-Fan Hsu, Jeng-Lin Li
Abstract
Vision foundation models have enabled strong training-free anomaly detection (AD). However, most existing approaches rely primarily on independent local patch features, leaving the global contextual information encoded by Vision Transformers (ViTs) underexploited. In this work, we identify the dual characteristics of the ViT [CLS] token: its embedding provides anomaly-invariant global semantic representation, while its attention maps implicitly highlight spatially abnormal regions. Building on this observation, we propose a fully automated AD framework leveraging global context to remove manual tunings. Our framework introduces (1) an automatic augmentation selection strategy driven by [CLS]-level semantic consistency, and (2) an attention-guided feature reweighting mechanism that dynamically adjusts patch contributions according to [CLS] attention saliency. By integrating these components over multi-level features, our method achieves stable anomaly scoring and precise localization without training or parameter tuning. Under the one-shot setting, it achieves Image-AUC scores of 97.7%, 93.2%, and 84.5% on MVTec-AD, VisA, and Real-IAD. Using a single fixed configuration across categories, backbones, and datasets, the method establishes a new state-of-the-art for plug-and-play, training-free anomaly detection while maintaining strong robustness and practical scalability.