We present a new framework for Patch Distribution Modeling, PaDiM, to\nconcurrently detect and localize anomalies in images in a one-class learning\nsetting. PaDiM makes use of a pretrained convolutional neural network (CNN) for\npatch embedding, and of multivariate Gaussian distributions to get a\nprobabilistic representation of the normal class. It also exploits correlations\nbetween the different semantic levels of CNN to better localize anomalies.\nPaDiM outperforms current state-of-the-art approaches for both anomaly\ndetection and localization on the MVTec AD and STC datasets. To match\nreal-world visual industrial inspection, we extend the evaluation protocol to\nassess performance of anomaly localization algorithms on non-aligned dataset.\nThe state-of-the-art performance and low complexity of PaDiM make it a good\ncandidate for many industrial applications.\n