Abstract: Non-destructive testing based on radiographic images is the key to quality control of welded steel pipes, while the high annotation cost and poor generalization to unknown defects limit supervised methods. Existing unsupervised approaches mainly rely on reconstruction errors or single feature modeling, which often suffer from missed detections under complex backgrounds and low-contrast defects. To address these issues, this paper proposes an intra and inter guided reverse distillation framework for unsupervised radiographic weld defect detection. By introducing intra and inter correlation learning mechanisms as prototypes, the proposed method enhances the discrimination of subtle defects. Experimental results show that the proposed approach outperforms state-of-the-art unsupervised methods, demonstrating its effectiveness for automated weld inspection.