Skip to content

This is an unofficial implementation of the paper “PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization”.

License

Notifications You must be signed in to change notification settings

xiahaifeng1995/PaDiM-Anomaly-Detection-Localization-master

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PaDiM-Anomaly-Detection-Localization-master

This is an implementation of the paper PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization.

This code is heavily borrowed from both SPADE-pytorch(https://github.com/byungjae89/SPADE-pytorch) and MahalanobisAD-pytorch(https://github.com/byungjae89/MahalanobisAD-pytorch) projects

Requirement

  • python == 3.7
  • pytorch == 1.5
  • tqdm
  • sklearn
  • matplotlib

Datasets

MVTec AD datasets : Download from MVTec website

Results

Implementation results on MVTec

  • Image-level anomaly detection accuracy (ROCAUC)
MvTec R18-Rd100 WR50-Rd550
Carpet 0.984 0.999
Grid 0.898 0.957
Leather 0.988 1.0
Tile 0.959 0.974
Wood 0.990 0.988
All texture classes 0.964 0.984
Bottle 0.996 0.998
Cable 0.855 0.922
Capsule 0.870 0.915
Hazelnut 0.841 0.933
Metal nut 0.974 0.992
Pill 0.869 0.944
Screw 0.745 0.844
Toothbrush 0.947 0.972
Transistor 0.925 0.978
Zipper 0.741 0.909
All object classes 0.876 0.941
All classes 0.905 0.955
  • Pixel-level anomaly detection accuracy (ROCAUC)
MvTec R18-Rd100 WR50-Rd550
Carpet 0.988 0.990
Grid 0.936 0.965
Leather 0.990 0.989
Tile 0.917 0.939
Wood 0.940 0.941
All texture classes 0.953 0.965
Bottle 0.981 0.982
Cable 0.949 0.968
Capsule 0.982 0.986
Hazelnut 0.979 0.979
Metal nut 0.967 0.971
Pill 0.946 0.961
Screw 0.972 0.983
Toothbrush 0.986 0.987
Transistor 0.968 0.975
Zipper 0.976 0.984
All object classes 0.971 0.978
All classes 0.965 0.973

ROC Curve

  • ResNet18

  • Wide_ResNet50_2

Localization examples

Reference

[1] Thomas Defard, Aleksandr Setkov, Angelique Loesch, Romaric Audigier. PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization. https://arxiv.org/pdf/2011.08785

[2] https://github.com/byungjae89/SPADE-pytorch

[3] https://github.com/byungjae89/MahalanobisAD-pytorch

About

This is an unofficial implementation of the paper “PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization”.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages