UOOD-MCD

This is an pytorch implementation of Unsupervised Out-of-Distribution Detection by Maximum Classifier Discrepancy.

Requirements

• Python 3.7
• PyTorch 1.1.0
• torchvision 0.3.0
• progress
• matplotlib
• numpy

Preparation

Download five out-of-distributin datasets provided by ODIN:

• Tiny-ImageNet (crop)
• Tiny-ImageNet (resize)
• LSUN (crop)
• LSUN (resize)
• iSUN

Here is an example code of downloading Tiny-ImageNet (crop) dataset. In the root directory, run

mkdir data
cd data
wget https://www.dropbox.com/s/avgm2u562itwpkl/Imagenet.tar.gz
tar -xvzf Imagenet.tar.gz
cd ..

Usage

Download Pre-trained Models on CIFAR

We provide download links of cifar10/100 pre-trained models.

• Pretrained Models

In the root directory, run

mkdir pretrained
cd pretrained
wget https://www.dropbox.com/s/qjitycxijexzp8y/pretrained.zip
unzip pretrained.zip
cd ..

Train single model

Finetune DenseNet on CIFAR-10 as ID and TinyImageNet as OOD.

python train.py -c checkpoints/cifar10_Imagenet_ckpt --gpu 0 --resume pretrained/cifar10_dense.pth.tar --out-dataset Imagenet

Trained model will be saved at checkpoints/cifar10_Imagenet_ckpt.

Train all models

python train_all.py --gpu 0

This script will finetune models of DenseNet/WideResNet on CIFAR-10/100 as ID and five other datasets as OOD which results in 20 models. Trained model will be saved at checkpoints.

Test

For example, to test DenseNet-BC trained on CIFAR-10 where TinyImageNet (crop) is the out-of-distribution dataset, please run

python test.py --result checkpoints/cifar10_Imagenet_ckpt

References

• [1]: Q. Yu and K. Aizawa. "Unsupervised Out-of-Distribution Detection by Maximum Classifier Discrepancy", in ICCV, 2019.
• [2]: S. Liang, Y. Li and R. Srikant. "Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks", in ICLR, 2018.