PyTorch re-implementation of PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization
| category | Paper | My Implementation |
|---|---|---|
| zipper | 0.985 | 0.923 |
| wood | 0.949 | 0.992 |
| transistor | 0.975 | 0.998 |
| toothbrush | 0.988 | 0.883 |
| tile | 0.941 | 0.994 |
| screw | 0.985 | 0.815 |
| pill | 0.957 | 0.958 |
| metal_nut | 0.972 | 0.992 |
| leather | 0.992 | 1.000 |
| hazelnut | 0.982 | 0.985 |
| grid | 0.973 | 0.959 |
| carpet | 0.991 | 0.997 |
| capsule | 0.985 | 0.937 |
| cable | 0.967 | 0.930 |
| bottle | 0.983 | 1.000 |
| category | Paper | My Implementation |
|---|---|---|
| zipper | 0.959 | 0.935 |
| wood | 0.911 | 0.891 |
| transistor | 0.845 | 0.949 |
| toothbrush | 0.931 | 0.915 |
| tile | 0.860 | 0.826 |
| screw | 0.944 | 0.936 |
| pill | 0.927 | 0.952 |
| metal_nut | 0.856 | 0.933 |
| leather | 0.978 | 0.978 |
| hazelnut | 0.926 | 0.937 |
| grid | 0.946 | 0.866 |
| carpet | 0.962 | 0.952 |
| capsule | 0.935 | 0.921 |
| cable | 0.888 | 0.918 |
| bottle | 0.948 | 0.951 |
- CUDA 10.2
- nvidia-docker2
a) Download docker image and run docker container
docker pull taikiinoue45/mvtec:padim
docker run --runtime nvidia -it taikiinoue45/mvtec:padim /bin/bash
b) Run Experiment
python run.py ./config.yaml params.category=bottle params.tracking_uri=file:///app/PaDiM/mlruns
- github: https://github.com/taikiinoue45/
- twitter: https://twitter.com/taikiinoue45/
- linkedin: https://www.linkedin.com/in/taikiinoue45/