Paper Title
Pixel Level Anomalous Segmentation in Industrial Manufacturing Application

Abstract
Anomalous segmentation of pharmaceutical products is a crucial task among many pharmaceutical manufacturers and often requires a constant inspection since defective products would have an adverse consequence on the public healthcare. Thus, pharmaceutical products are often imposed by strict safety policing and undergo scrupulous product testing to meet the acceptable quality standards. With the continuous need to reduce production costs, the recognition and immediate alteration of imperfections from desired operating conditions become more significant. Timely detection and segmentation of fault can potentially avoid the progression of oddment and thus reduce productivity loss without distressing the quality and safety of the products. Moreover, an automated fault segmentation system narrows the reliance on human operators, which according to industrial statistics results in 70% of industrial accidents. Thus, this brings out to the current research which is formulated as an unsupervised learning problem. The aim of this paper is to address anomalous object segmentation, which considers any previously-unseen object category as anomalous on a challenging MVTec AD dataset employed for a category of capsules. Establishing an in-depth method analysis, based on pixel-wise performance metrics that are insensitive to object sizes, we empirically evaluate multiple state-of-the-art baseline methods, including several models specifically designed for calculating anomaly score and addressing anomaly segmentation, on our dataset using the test suite. Our experiment has demonstrated that the performance on the dataset achieves the new state of the art, with a significant margin over previous works and segments the anomalies more accurately. Keywords - Anomaly Segmentation, Anomaly Score, Pharmaceutical Capsule, MVTecADDataset, Pixel-Wise Metric, Unsupervised Learning