Main Dependencies torch 1.1.0 torchvision 0.3.0 sklearn 0.20.3 numpy 1.19.5 matplotlib 3.0.3 cuda 10.1
How to run:
You can use following command:
python main.py --normal_digit 0 --n_epochs 1000 --batch_size 200 --auxiliary_digit 1 --latent_dim 128 --name cifar --gamma_p 0 --gamma_l 0.2 --k 1 --dataset CIFAR --dir /CIFAR0.2/summary//
to train an Anomaly-Aware BiGAN on CIFAR-10 dataset. The result will save as ./dir/name.csv
You can also use the following command:
bash bash_cifar.sh bash bash_fmnist.sh bash bash_mnist.sh
to run the .sh example file.
option choices: dataset =[CIFAR,F-MNIST,MNIST] gamma_l = [0.01,0.05,0.1,0.2] gamma_p= [0,0.01,0.05,0.1,0.2] k = [0,1,2,3,5] latent_dim = [128 (CIFAR), 100 (F-MNIST,MNIST)]
'main_unsupervised.py' is for the scenario Anoamly-Aware BiGAN without the leverage of collected anomalies (gamma_l=0 scenario). Use command: python main_unsupervised.py --normal_digit 0 --n_epochs 200 --batch_size 200 --latent_dim 100 --name fmnist --gamma_p 0 --dataset F-MNIST
to run.