Task-agnostic Out-of-Distribution Detection Using Kernel Density Estimation

PyTorch implementation of our paper Task-agnostic Out-of-Distribution Detection Using Kernel Density Estimation by Ertunc Erdil (email), Krishna Chaitanya, Neerav Karani, and Ender Konukoglu.

Citation

If you find this code helpful in your research please cite the following paper:

@article{erdil2021taskagnostic,
  title={Task-agnostic out-of-distribution detection using kernel density estimation},
  author={Erdil, Ertunc and Chaitanya, Krishna and Karani, Neerav and Konukoglu, Ender},
  journal={arXiv preprint arXiv:2006.10712},
  year={2021}
}

Install

conda create --name <env_name> # Create virtual environment - optional
conda install -c conda-forge scikit-learn
conda install numpy
conda install pytorch torchvision cudatoolkit=9.2 -c pytorch
pip install torchsummary

Downloading datasets and models

Download the datasets and copy to ./data folder

Download the pre-trained models and copy to ./pre_trained folder

How to run

python main_train.py config/cifar10/cifar10_kde_cfg_01.py # Running on CIFAR10 dataset

python main_train.py config/cifar100/cifar100_kde_cfg_01.py # Running on CIFAR100 dataset