Zero-Shot Out-Of-Distribution Detection with Outlier Label Exposure

By Choubo Ding, Guansong Pang

Official PyTorch implementation of the paper “Zero-Shot Out-Of-Distribution Detection with Outlier Label Exposure”

Code is modified from CLIPN.

Installation

git clone https://github.com/Choubo/OLE.git
cd OLE
conda create -n ole python=3.8
conda activate ole
pip install -r requirements.txt

Dataset Preparation

Our experiments utilize the following datasets:

In-distribution

• ImageNet-1K: The ILSVRC-2012 version, serving as our primary in-distribution dataset. It can be downloaded here

Out-of-distribution

We use subsampled versions of the following datasets, with classes overlapping ImageNet-1K removed:

• iNaturalist
• SUN
• Places
• Texture

For detailed download instructions and preprocessing steps, please refer to the MOS.

Quick Start

1. Download and place the checkpoint:

   • Download the checkpoint from this link.
   • Place the downloaded checkpoint in the src folder:

     mv path/to/downloaded/clipn_checkpoint.pth src/

2. Run outlier prototype learning and OLE evaluation:

   cd src
   bash run.sh

This script will execute the outlier prototype learning process and perform OLE evaluation.

Citation

If you use our codebase, please cite our work:

@inproceedings{
ding2024zero,
title={Zero-Shot Out-of-Distribution Detection with Outlier Label Exposure},
author={Choubo Ding and Guansong Pang},
booktitle={2024 International Joint Conference on Neural Networks},
year={2024}
}