A Real-time Unsupervised Two-stage Framework for Fabric Defect Detection

Fabric defect detection plays a vital role in ensuring product quality. In this paper, we proposed an effective and real-time unsupervised two-stage framework only trained with defect-free images for fabric defect detection. It consists of a lightweight reconstruction network for repairing and reconstructing the fabric image, and a following lightweight decision network for accurate defect localization. Experiments on three fabric datasets evaluated the effectiveness of the proposed method, which achieves excellent performance in detection accuracy (98.34% F1 score), model size (162.6K), model computational complexity (146.48M floating-point operations), and detection speed (122.59 frames per second). Our proposed two-stage framework can be applied to inspect fabric on a practical production line with excellent accuracy and high detection speed.