-
Notifications
You must be signed in to change notification settings - Fork 22
home
A set of functions and classes for performing anomaly detection in images using features from pretrained neural networks.
The package includes functions and classes for extracting, modifying and comparing features. It also includes unofficial implementations of PaDiM and PatchCore.
Some code has been borrowed and/or inspired by other repositories, see code reference below.
See wiki for documentation.
Clone the repository
git clone https://github.com/OpenAOI/anodet.git
Install the package
cd anodet
python -m pip install -r requirements.txt
python -m pip install .
# Prepare a dataloader and fit a model to the data
dataloader = DataLoader(...)
padim = anodet.Padim()
padim.fit(dataloader)
# Prepare some test images and make predictions
batch = ...
image_scores, score_map = padim.predict(batch)
See notebooks for in depth examples.
Install the package in editable mode
python -m pip install --editable [PATH TO REPOSITORY]
Install packages for testing
python -m pip install pytest pytest-mypy pytest-flake8
Run tests
cd [PATH TO REPOSITORY]
pytest --mypy --flake8
For configuration of pytest, mypy and flake8 edit setup.cfg.
Install pydoc-markdown
python -m pip install pydoc-markdown
Clone docs repository
git clone https://github.com/OpenAOI/anodet.wiki.git
Run script
cd anodet.wiki
python generate_docs.py --source-path=[PATH TO REPOSITORY] --package-name="anodet" --save-path=.
Some parts used in patch_core.py : https://github.com/hcw-00/PatchCore_anomaly_detection
Code in directory sampling_methods : https://github.com/google/active-learning
concatenate_two_layers function in feature_extraction.py : https://github.com/xiahaifeng1995/PaDiM-Anomaly-Detection-Localization-master
pytorch_cov function in utils.py : https://github.com/pytorch/pytorch/issues/19037