Student-Teacher anomaly detection

This is an implementation of the paper Uninformed Students: Student–Teacher Anomaly Detection with Discriminative Latent Embeddings.

How to use

• Download a dataset from MVTec website and extract it under the /data folder. Let us download the carpet dataset for the example. You might need to run

chmod -R u+rw data

• Run the mvtec_dataset.sh script to prepare the dataset in the correct format.

./mvtec_dataset.sh carpet

• (Optional) Run resnet18_training.py script to train resnet18 further on your dataset

cd src
python3 resnet18_training.py carpet

• Run teacher_training.py to distil the knowledge of resnet18 on a smaller neural network. This will speed up the processing of images. This neural network, called the Teacher, outputs a 512-dimensional description vector for each patch of size 65x65 of the image

cd src
python3 teacher_training.py carpet

• Run students_training.py to train a set of M=3 students against the teacher network. The training of the students is done on an anomaly-free dataset. We expect them to generalize poorly in images containing anomalies

cd src
python3 students_training.py carpet

• Run anomaly_detection.py to obtain an anomaly map for each image of the test set. An anomaly map is computed using the variance of Students predictions and the error between Students predictions and Teacher.

cd src
python3 anomaly_detection.py carpet

Results

And more results are available under /result folder

Expected folder structure

├── data
│   ├── carpet
│   └── hazelnut
├── docs
│   ├── 9245_FastCNNFeature_BMVC.pdf
│   ├── anomaly_detection.pdf
│   └── anomaly_detection_summary.pdf
├── model
│   ├── carpet
│   └── hazelnut
├── mvtec_dataset.py
├── mvtec_dataset.sh
├── README.md
├── results
│   ├── anomaly_carpet_res1.png
│   ├── anomaly_carpet_res2.png
│   ├── anomaly_carpet_res3.png
│   ├── anomaly_carpet_res4.png
│   ├── anomaly_hazelnul_res2.png
│   └── anomaly_hazelnut_res1.png
└── src
├── AnomalyDataset.py
├── anomaly_detection.py
├── AnomalyNet.py
├── AnomalyResnet18.py
├── ExtendedAnomalyNet.py
├── FDFEAnomalyNet.py
├── FDFE.py
├── resnet18_training.py
├── students_training.py
├── teacher_training.py
└── utils.py

References

Original paper

• https://arxiv.org/pdf/1911.02357v2.pdf

Dataset

• https://www.mvtec.com/fileadmin/Redaktion/mvtec.com/company/research/mvtec_ad.pdf

Fast Dense Feature Extraction

• https://www.dfki.de/fileadmin/user_upload/import/9245_FastCNNFeature_BMVC.pdf
• https://github.com/erezposner/Fast_Dense_Feature_Extraction