Abstract: Deep learning has been widely used in recent years for surface defect detection because of its excellent performance. However, current deep-learning-based approaches still remain a ...