Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping
This addresses anomaly detection in industrial settings by improving efficiency and accuracy, though it is incremental as it builds on existing multimodal methods.
The paper tackles industrial anomaly detection using point clouds and RGB images by introducing a framework that maps features between modalities on normal samples and detects anomalies from inconsistencies, achieving state-of-the-art performance on the MVTec 3D-AD dataset with faster inference and lower memory usage.
The paper explores the industrial multimodal Anomaly Detection (AD) task, which exploits point clouds and RGB images to localize anomalies. We introduce a novel light and fast framework that learns to map features from one modality to the other on nominal samples. At test time, anomalies are detected by pinpointing inconsistencies between observed and mapped features. Extensive experiments show that our approach achieves state-of-the-art detection and segmentation performance in both the standard and few-shot settings on the MVTec 3D-AD dataset while achieving faster inference and occupying less memory than previous multimodal AD methods. Moreover, we propose a layer-pruning technique to improve memory and time efficiency with a marginal sacrifice in performance.