ConceptADapt: Concept-guided Adaptive Feature Reconstruction with Dynamic Attention for Few-Shot Industrial Anomaly Detection

2026-08-06

AI Summary

This work addresses the challenging problem of few-shot anomaly detection and localization in industrial cold-start scenarios, where only a handful of normal samples are available. To tackle this, the authors propose a lightweight, adaptive reconstruction framework that leverages features from a pretrained foundation model. The approach employs a concept-guided feature recalibration strategy, integrating a dynamic attention mechanism with a sparse autoencoder to effectively mitigate shortcut learning in feature representations. Furthermore, it incorporates LoRA (Low-Rank Adaptation) to enable efficient fine-tuning and rapid inference adaptation. Extensive experiments on three major benchmarks—MVTec-AD, VisA, and MPDD—demonstrate that the proposed method consistently outperforms existing state-of-the-art approaches across various shot settings, exhibiting strong generalization capability and robustness.

Abstract

Few-shot industrial anomaly detection (FS-IAD) focuses on detecting and localizing visual defects in industrial inspection during the cold-start phase, where only a limited number of normal training samples are available per category. Recent advances in this field predominantly leverage visual features from foundation-model and have achieved promising performance. Despite the strong representational power of foundation-model features, the model generalization remains fragile due to the extreme scarcity of normal training data.To address this pivotal issue, we propose ConceptADapt, a concept-guided adaptive feature reconstruction model with dynamic attention. Specifically, our model pre-learns a set of fixed normal concepts from the limited support features and leverages them to mine relationships with query features, thereby recalibrating their statistics for improved anomaly detection at test time. To mitigate the prevalent feature shortcut problem, which is particularly severe under low-data regimes, we further develop a dynamic attention mechanism integrated with sparse autoencoders to learn robust normal concepts during training. Moreover, to enable fast adaptation during inference, our model remains lightweight by incorporating LoRA into the attention module, which introduces only minimal updating parameters.Extensive experiments on three widely adopted FS-IAD benchmarks, including MVTec-AD, VisA, and MPDD, demonstrate that our model consistently outperforms state-of-the-art (SOTA) approaches across both detection and localization tasks, achieving significant improvements under various shot settings.

Problem

• Few-Shot Industrial Anomaly Detection
• Cold-Start Phase
• Limited Normal Samples
• Visual Defect Detection
• Industrial Inspection

Innovation

• concept-guided
• dynamic attention
• few-shot anomaly detection
• feature reconstruction
• LoRA