This is a fork of the OpenOOD repository, containing additions to reproduce experiments made in the paper Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection.
For installation, please refer to the original repository.
Additions include CADet postprocessor, GMM ensemble postprocessor, and the store_stats pipeline which is used to compute and store scores, as well as a number of bug fixes (the repository now correctly supports ViT) and some additional features.
Examples of config files are provided in configs/postprocessors/cadet_postprocessor.yml and configs/postprocessors/gmm_ens_postprocessor.yml.
For the gmm ensemble postprocessor, first make sure all scores have been saved in config.stats_dir/network_name/dataset_name/statistic_name.
The model will be trained on the scores provided for the dataset config.postprocessor.postprocessor_args.id_ds_name and will compute stats for the datasets in config.ood_dataset.ood.datasets.