arXiv:2606.14129v1 Announce Type: new 
Abstract: Reconstruction-based anomaly detection is attractive for industrial inspection, but scaling it from category-specific training to a one-for-all setting is challenging. A single model must reconstruct diverse normal appearances without copying abnormal details, which exposes two coupled failure modes: identical shortcut, where anomalies pass through the reconstruction path, and mis-reconstruction, where normal categories are confused with one another. We propose \textbf{BoRAD}, a label-free training framework that treats this as a representation-capacity allocation problem. BoRAD uses a shared learnable prototype bank to impose two complementary regularizers: spatial prototype alignment contracts local within-prototype variation to suppress anomaly copying, while prototype-relative global alignment preserves between-prototype structure and improves sensitivity to abnormal angular deviations. The prototype bank and prediction heads are used only during training; inference remains a standard teacher-student feature discrepancy pass, with no class labels, negative pairs, memory retrieval, or prototype lookup. BoRAD achieves competitive one-for-all anomaly detection performance, including 86.2\% mAD on MVTec AD, 80.7\% mAD on VisA and 73.1\% mAD on Real-IAD. Diagnostic analyses further show reduced anomaly leakage, improved normal-category separability, and stronger anomaly-normal score separation.

تم تقديم إطار عمل جديد يسمى BoRAD للكشف عن الشذوذ متعدد الفئات، حيث يتناول التحديات المتعلقة بتوسيع نطاق الكشف عن الشذوذ القائم على إعادة البناء من التدريب المحدد للفئة إلى نموذج شامل. يستخدم BoRAD بنك بروتوتيب قابل للتعلم مشترك لتعزيز قدرة النموذج على التمييز بين المظاهر الطبيعية وغير الطبيعية مع تقليل إعادة البناء الخاطئة ونسخ الشذوذ.

Se ha introducido un nuevo marco llamado BoRAD para la detección de anomalías multiclasificación, abordando los desafíos de escalar la detección de anomalías basada en la reconstrucción desde un entrenamiento específico de categoría a un modelo universal. BoRAD emplea un banco de prototipos aprendibles compartidos para mejorar la capacidad del modelo para diferenciar entre apariencias normales y anormales, minimizando la mala reconstrucción y la copia de anomalías.

Un nouveau cadre appelé BoRAD a été introduit pour la détection d'anomalies multi-classes, abordant les défis de l'échelle de la détection d'anomalies basée sur la reconstruction, passant d'un entraînement spécifique à une catégorie à un modèle universel. BoRAD utilise une banque de prototypes apprenables partagée pour améliorer la capacité du modèle à différencier les apparences normales et anormales tout en minimisant la mauvaise reconstruction et la copie d'anomalies.

A new framework called BoRAD has been introduced for multi-class anomaly detection, addressing the challenges of scaling reconstruction-based anomaly detection from category-specific training to a universal model. BoRAD employs a shared learnable prototype bank to enhance the model's ability to differentiate between normal and abnormal appearances while minimizing mis-reconstruction and anomaly copying.