A StyleGAN3-Based Data Augmentation Method for Ceramic Defect Detection

Abstract

Deep learning is currently the mainstream method for ceramic defect detection, and it requires a large number of defect samples to train the network. However, collecting these defect samples is very time-consuming and deep learning suffers from few-shot learning problems. In this study, a StyleGAN3-based data augmentation method for ceramic defect detection was proposed which can generate ceramic defect samples and thus reduce the data collection work. Experiments show that our method uses less training time, has a more stable training process, and can improve the accuracy of the detection network.