基于扩散合成与特征挖掘的工业图像异常检测

卜玉真,余家斌,马道滨,陈梁宇,孙龙,杨力,章东平

Industrial image anomaly detection via diffusion synthesis and feature mining

Yuzhen BU,Jiabin YU,Daobin MA,Liangyu CHEN,Long SUN,Li YANG,Dongping ZHANG

表 5 不同异常检测方法在MPDD数据集上的整体性能对比

Tab.5 Overall performance comparison of different anomaly detection methods on MPDD dataset

方法 Image AUROC/% Pixel AUROC/% PatchCore 82.1 95.7 CFlow 86.1 97.7 PaDiM 74.8 96.7 SPADE 77.1 95.9 DAGAN 72.5 83.3 Skip-GANomaly 64.8 82.2 RealNet 96.3 98.2 本研究算法 98.4 98.7