qkqkfldis1/GAN-Ensemble-for-Anomaly-Detection

This repository is the PyTorch implementation of GAN Ensemble for Anomaly Detection.

GAN-Ensemble-for-Anomaly-Detection

This repository contains PyTorch implementation of the following paper: GAN-Ensemble-for-Anomaly-Detection

0. Environment Setup

pip install -r requirements.txt

1. Table of Contents

• GAN-Ensemble-for-Anomaly-Detection

  • 1. Table of Contents

  • 2. Experiment

  • 3. Training

2. Experiment

To replicate the results in the paper for MNIST and CIFAR10 dataset, run the following commands:

#MNIST
sh experiments/run_mnist_en_fanogan.sh
sh experiments/run_mnist_en_egbad.sh
# CIFAR
sh experiments/run_cifar_en_fanogan.sh
sh experiments/run_cifar_en_egbad.sh
#OCT
sh experiments/run_oct_en_fanogan.sh
#KDD99
sh experiments/run_oct_en_egbad.sh

3. Training

To list the arguments, run the following command:

python train.py -h

To train the model on MNIST dataset for a given anomaly class, run the following:

python train.py \
    --dataset mnist                                                                \
    --niter <number-of-epochs>                                                     \
    --abnormal_class  <0,1,2,3,4,5,6,7,8,9>                                        \
    --setting <model-name: f-anogan, egbad, ganomaly, skipgan>                     \
    --n_G <number of ensemble generators>                                          \
    --n_D <number of ensemble discriminators>                                      \

To train the model on CIFAR10 dataset for a given anomaly class, run the following:

python train.py \
    --dataset cifar10                                                             \
    --niter <number-of-epochs>                                                    \
    --abnormal_class                                                              \
        <0-9 for :airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck>    \
    --setting <model-name: f-anogan, egbad, ganomaly, skipgan>                     \
    --n_G <number of ensemble generators>                                          \
    --n_D <number of ensemble discriminators>                                      \
         

To train the model on OCT dataset for a given anomaly class, run the following:

python train.py \
    --dataset OCT                                                                  \
    --niter <number-of-epochs>                                                     \
    --setting <model-name: f-anogan, egbad, ganomaly, skipgan>                     \
    --n_G <number of ensemble generators>                                          \
    --n_D <number of ensemble discriminators>                                      \
         

To train the model on KDD99 dataset for a given anomaly class, run the following:

python train.py \
    --dataset KDD99                                                                \
    --niter <number-of-epochs>                                                     \
    --setting <model-name: f-anogan, egbad>                                        \
    --n_G <number of ensemble generators>                                          \
    --n_D <number of ensemble discriminators>                                      \