PhD Dissertation Defense by LIU Zhonghang | Towards Real-world Unsupervised Anomaly Detection for Images

Towards Real-world Unsupervised Anomaly Detection for Images

ABOUT THE TALK

In the era of big data, data quality plays a critical role in computer vision. If the training dataset is contaminated, the model’s performance might accordingly decrease. To address this challenge, unsupervised anomaly detection (UAD), which automatically removes these anomalous data points has become an attractive research area. 

We begin by introducing LVAD, a novel statistical embedding-based approach, establishing a strong baseline for unsupervised anomaly scoring. Besides, to bridge the gap between anomaly scoring and thresholding (label prediction), we propose Multi-T. Unlike previous threshold learners that depend on a single global threshold, Multi-T introduces the concept of multiple thresholds. Multi-T transforms the unlabeled target dataset into a weakly supervised resource, allowing significant improvements to existing anomaly scoring methods. 

Despite these above advances, a key challenge remains: the instability of UAD methods under varying contamination levels (anomaly percentages) in target datasets. To address this, we introduce FlexUAD, a training-free and plug-and-play framework that incorporates a contamination factor estimator. FlexUAD adaptively selects the appropriate anomaly detector based on the estimated contamination factor, ensuring both high efficacy and stability across diverse settings. 

In summary, this dissertation contributes a cohesive suite of methods: LVAD for robust anomaly scoring, Multi-T for multiple thresholding, and FlexUAD for contamination factor estimation that collectively push the boundaries of unsupervised image anomaly detection towards greater stability and efficiency. Additionally, we will further discuss some industrial applications of UAD, i.e., industrial inspection.

SPEAKER BIOGRAPHY

LIU Zhonghang is a Ph.D. candidate in Computer Science at the School of Computing and Information Systems, Singapore Management University (SMU), under the supervision of Dr. Daniel LIN. He mainly focuses on unsupervised image anomaly detection via high-dimensional space analysis. Besides, he is interested in improving the detection accuracy and efficiency of some real-world problems, such as industrial inspection. He enjoys reading and doing sports in his leisure time.