• multivariate Gaussian distribution N (μij , Σij )

1. score patch embedding xij by the Mahalanobis distance
2. PaDiM keeps large amount of matrices (Σki)^−1, i ∈ {Hk ×Wk } for Mahalanobis distance.
   1. PaDiM has been inspired by Rippel et al. [29] who firstly advocated to use this measure for anomaly detection without localization.
   2. distribution per location, too dependent on matched locations

References

1. PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization, icprw21
2. [29] Modeling the Distribution of Normal Data in Pre-Trained Deep Features for Anomaly Detection, 20