Abstract: Recently, local-editing-based transformations are introduced in anomaly detection for defect visual inspection, which construct a pretext task with the paradigm of self-supervised learning.