In this work, we present a simple yet effective CNN-based framework for segmentation and dectection designed for anomaly detection, focusing on both structural and logical anomalies by inspecting the visual surface. Our model is engineered for rapid training while maintaining high detection accuracy, addressing the challenges of single-class and few-shot learning. Evaluations on five different classes of the MVTec LOCO dataset yielded an impressive average image score of 0.7786, highlighting the model's robustness and reliability across diverse anomaly detection scenarios. This demonstrates that even with a simple and fast approach, high performance in anomaly detection can be achieved.
Comments