https://github.com/ServiceNow/broad-openood

https://github.com/ServiceNow/broad-openood

Last synced: 27 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/ServiceNow/broad-openood
• Owner: ServiceNow
• License: mit
• Created: 2023-08-22T10:59:32.000Z (about 1 year ago)
• Default Branch: master
• Last Pushed: 2023-08-22T11:04:29.000Z (about 1 year ago)
• Last Synced: 2024-07-22T05:19:29.510Z (4 months ago)
• Language: Python
• Size: 363 KB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
This is a fork of the [`OpenOOD`](https://github.com/Jingkang50/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](https://github.com/Jingkang50/OpenOOD).

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`.