GANomaly

This repository contains PyTorch implementation of the following paper: GANomaly: Semi-Supervised Anomaly Detection via Adversarial Training [1]

1. Table of Contents

• GANomaly

• Training on MNIST

• Training on CIFAR10

• Train on Custom Dataset

• Table of Contents

• Installation

• Experiment

• Training

• Citing GANomaly

• Reference

2. Installation

1. First clone the repository

   git clone https://github.com/samet-akcay/ganomaly.git

2. Create the virtual environment via conda

   conda create -n ganomaly python=3.7

3. Activate the virtual environment.

   conda activate ganomaly

4. Install the dependencies.

   pip install --user --requirement requirements.txt

3. Experiment

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

# MNISTsh experiments/run_mnist.sh# CIFARsh experiments/run_cifar.sh # CIFAR10

4. Training

To list the arguments, run the following command:

python train.py -h

4.1. Training on MNIST

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>  
    --display                               # optional if you want to visualize

4.2. Training on CIFAR10

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                                                    
        <plane, car, bird, cat, deer, dog, frog, horse, ship, truck>    
    --display                       # optional if you want to visualize

4.3. Train on Custom Dataset

To train the model on a custom dataset, the dataset should be copied into ./data directory, and should have the following directory & file structure:

Custom Dataset
├── test
│   ├── 0.normal
│   │   └── normal_tst_img_0.png
│   │   └── normal_tst_img_1.png
│   │   ...
│   │   └── normal_tst_img_n.png
│   ├── 1.abnormal
│   │   └── abnormal_tst_img_0.png
│   │   └── abnormal_tst_img_1.png
│   │   ...
│   │   └── abnormal_tst_img_m.png
├── train
│   ├── 0.normal
│   │   └── normal_tst_img_0.png
│   │   └── normal_tst_img_1.png
│   │   ...
│   │   └── normal_tst_img_t.png

Then model training is the same as training MNIST or CIFAR10 datasets explained above.

python train.py                     
    --dataset <name-of-the-data>    
    --isize <image-size>            
    --niter <number-of-epochs>      
    --display                       # optional if you want to visualize

For more training options, run python train.py -h.

5. Citing GANomaly

If you use this repository or would like to refer the paper, please use the following BibTeX entry

@inproceedings{akcay2018ganomaly,
  title={Ganomaly: Semi-supervised anomaly detection via adversarial training},
  author={Akcay, Samet and Atapour-Abarghouei, Amir and Breckon, Toby P},
  booktitle={Asian Conference on Computer Vision},
  pages={622--637},
  year={2018},
  organization={Springer}
}

6. Reference

[1]  Akcay S., Atapour-Abarghouei A., Breckon T.P. (2019) GANomaly: Semi-supervised Anomaly Detection via Adversarial Training. In: Jawahar C., Li H., Mori G., Schindler K. (eds) Computer Vision – ACCV 2018. ACCV 2018. Lecture Notes in Computer Science, vol 11363. Springer, Cham