This is an unofficial implementation of the paper Towards Total Recall in Industrial Anomaly Detection.

We measured accuracy and speed for percentage_coreset=0.01, 0.1 and 0.25.

This code was implimented with patchcore-inspection, thanks.

Some "ex" are explainability, express delivery, flexibility, extra algolithm and so on.

• faiss-gpu (easy to install with conda : ref)
• torch
• torchvision
• numpy
• opencv-python
• scipy
• argparse
• matplotlib
• scikit-learn
• torchinfo
• tqdm

Install prerequisites with:

conda install --file requirements.txt

Please download MVTec AD dataset.

After downloading, place the data as follows:

./
├── main.py
└── mvtec_anomaly_detection
    ├── bottle
    │   ├── ground_truth  # pixel annotation of test data other than good
    │   │   ├── broken_large
    │   │   ├── broken_small
    │   │   └── contamination
    │   ├── test  # other than good is anomaly data
    │   │   ├── broken_large
    │   │   ├── broken_small
    │   │   ├── contamination
    │   │   └── good
    │   └── train  # good only
    │       └── good
    ├── cable
    ├── cable
    ├── capsule
    ├── carpet
    ├── grid
    ├── hazelnut
    ├── leather
    ├── metal_nut
    ├── pill
    ├── screw
    ├── tile
    ├── toothbrush
    ├── transistor
    ├── wood
    └── zipper

When using custom dataset, place the data as follows:

./
├── main.py
└── custom_data
    └── theme
        ├── ground_truth  # pixel annotation of test data other than good
        │   ├── anomaly_type_a
        │   └── anomaly_type_b
        ├── test  # other than good is anomaly data
        │   ├── anomaly_type_a
        │   ├── anomaly_type_b
        │   └── good
        └── train  # good only
            └── good

To test PatchCore on MVTec AD dataset:

python main.py

After running the code above, you can see the ROCAUC results in result/roc_curve.png

To test PatchCore on custom dataset:

python main.py --path_data ./custom_data

Below is the implementation result of the test set ROCAUC on the MVTec AD dataset.

CPU : Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz
GPU : Tesla V100 SXM2

• percentage_coreset = 0.01

• percentage_coreset = 0.1

• percentage_coreset = 0.25

• bottle

• cable

• capsule

• carpet

• grid

• hazelnut

• leather

• metal_nut

• pill

• screw

• tile

• toothbrush

• transistor

• wood

• zipper

• bottle (test case : broken_large)

• cable (test case : bent_wire)

• capsule (test case : crack)

• carpet (test case : color)

• grid (test case : bent)

• hazelnut (test case : crack)

• leather (test case : color)

• metal_nut (test case : bent)

• pill (test case : color)

• screw (test case : manipulated_front)

• tile (test case : crack)

• toothbrush (test case : defective)

• transistor (test case : bent_lead)

• wood (test case : color)

• zipper (test case : broken_teeth)

We also implement a similar algorithm, SPADE.
https://github.com/any-tech/SPADE-fast/tree/main

There is an explanatory article.
https://tech.anytech.co.jp/entry/2023/03/24/100000