PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization

17 Nov 2020  ·  Thomas Defard, Aleksandr Setkov, Angelique Loesch, Romaric Audigier ·

We present a new framework for Patch Distribution Modeling, PaDiM, to concurrently detect and localize anomalies in images in a one-class learning setting. PaDiM makes use of a pretrained convolutional neural network (CNN) for patch embedding, and of multivariate Gaussian distributions to get a probabilistic representation of the normal class. It also exploits correlations between the different semantic levels of CNN to better localize anomalies. PaDiM outperforms current state-of-the-art approaches for both anomaly detection and localization on the MVTec AD and STC datasets. To match real-world visual industrial inspection, we extend the evaluation protocol to assess performance of anomaly localization algorithms on non-aligned dataset. The state-of-the-art performance and low complexity of PaDiM make it a good candidate for many industrial applications.

PDF Abstract

Results from the Paper

 Ranked #1 on on MVTecAD

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Anomaly Detection Hyper-Kvasir Dataset PaDiM AUC 0.923 # 4
Anomaly Detection LAG PaDiM AUC 0.688 # 5
MVTecAD Diadem Absolute Time (ms) 18 ms # 1
Anomaly Detection MVTec AD PaDiM-WR50-Rd550 Detection AUROC 95.3 # 62
Segmentation AUROC 97.5 # 45
FPS 4.4 # 17
Anomaly Detection MVTec AD PaDiM-R18-Rd100 Segmentation AUROC 96.7 # 56
Anomaly Detection MVTec AD PaDiM-EfficientNet-B5 Detection AUROC 97.9 # 47
Anomaly Detection VisA PaDiM Segmentation AUPRO (until 30% FPR) 85.9 # 12