Modifying DeepSVDD for to work on the MRNet dataset.

Citation

You find a PDF of the Deep One-Class Classification ICML 2018 paper at http://proceedings.mlr.press/v80/ruff18a.html.

If you use our work, please also cite the paper:

@InProceedings{pmlr-v80-ruff18a,
  title     = {Deep One-Class Classification},
  author    = {Ruff, Lukas and Vandermeulen, Robert A. and G{\"o}rnitz, Nico and Deecke, Lucas and Siddiqui, Shoaib A. and Binder, Alexander and M{\"u}ller, Emmanuel and Kloft, Marius},
  booktitle = {Proceedings of the 35th International Conference on Machine Learning},
  pages     = {4393--4402},
  year      = {2018},
  volume    = {80},
}

Installation

This code is written in Python 3.7 and requires the packages listed in requirements.txt.

Clone the repository to your local machine and directory of choice:

git clone https://github.com/lukasruff/Deep-SVDD-PyTorch.git

To run the code, we recommend setting up a virtual environment, e.g. using virtualenv or conda:

virtualenv

# pip install virtualenv
cd <path-to-Deep-SVDD-PyTorch-directory>
virtualenv myenv
source myenv/bin/activate
pip install -r requirements.txt

conda

cd <path-to-Deep-SVDD-PyTorch-directory>
conda create --name myenv
source activate myenv
while read requirement; do conda install -n myenv --yes $requirement; done < requirements.txt

Running experiments

We currently have implemented the MNIST (http://yann.lecun.com/exdb/mnist/) and CIFAR-10 (https://www.cs.toronto.edu/~kriz/cifar.html) datasets and simple LeNet-type networks.

Have a look into main.py for all possible arguments and options.

MNIST example

cd <path-to-Deep-SVDD-PyTorch-directory>

# activate virtual environment
source myenv/bin/activate  # or 'source activate myenv' for conda

# create folder for experimental output
mkdir log/mnist_test

# change to source directory
cd src

# run experiment
python main.py mnist mnist_LeNet ../log/mnist_test ../data --objective one-class --lr 0.0001 --n_epochs 150 --lr_milestone 50 --batch_size 200 --weight_decay 0.5e-6 --pretrain True --ae_lr 0.0001 --ae_n_epochs 150 --ae_lr_milestone 50 --ae_batch_size 200 --ae_weight_decay 0.5e-3 --normal_class 3;

This example trains a One-Class Deep SVDD model where digit 3 (--normal_class 3) is considered to be the normal class. Autoencoder pretraining is used for parameter initialization.

CIFAR-10 example

cd <path-to-Deep-SVDD-PyTorch-directory>

# activate virtual environment
source myenv/bin/activate  # or 'source activate myenv' for conda

# create folder for experimental output
mkdir log/cifar10_test

# change to source directory
cd src

# run experiment
python main.py cifar10 cifar10_LeNet ../log/cifar10_test ../data --objective one-class --lr 0.0001 --n_epochs 150 --lr_milestone 50 --batch_size 200 --weight_decay 0.5e-6 --pretrain True --ae_lr 0.0001 --ae_n_epochs 350 --ae_lr_milestone 250 --ae_batch_size 200 --ae_weight_decay 0.5e-6 --normal_class 3;

This example trains a One-Class Deep SVDD model where cats (--normal_class 3) is considered to be the normal class. Autoencoder pretraining is used for parameter initialization.

Examples

MNIST

Example of the 32 most normal (left) and 32 most anomalous (right) test set examples per class on MNIST according to Deep SVDD anomaly scores.

CIFAR-10

Example of the 32 most normal (left) and 32 most anomalous (right) test set examples per class on CIFAR-10 according to Deep SVDD anomaly scores.

License

MIT