sample command for training the model.

python train_dnn.py --dataset cifar10 --ood_dataset tiny-images --ood_train_x <path_to_ood_train_data>  --epochs 200 -use-gpu --nesterov --net_type wide-resnet --depth 40 --widen_factor 2 --dropout 0.3 --epochs 200 --seed 0 --expt_name cifar10_tiny --save_best_model --datadir /home/data/cifar

The trained model (saved using the --save_best_model above) is then evaluated on a test OoD data accoring to equation 2 in the paper.