Improving Out-of-Distribution Detection with Disentangled Foreground and Background Features

By Choubo Ding, Guansong Pang

Official PyTorch implementation of the paper “Improving Out-of-Distribution Detection with Disentangled Foreground and Background Features”. The paper is available at arXiv.

Code is modified from Google BiT and MOS.

Usage

1. Dataset Preparation

In-distribution dataset

Please download CIFAR10 and CIFAR100, then place them in ./dataset/cifar10 and ./dataset/cifar100, respectively.

Out-of-distribution dataset

We have curated 3 more OOD datasets from SVHN, Places, Textures.

1. SVHN:

   • Download from the official website.

2. Places:

   • Download from MOS or using the following command:

     wget http://pages.cs.wisc.edu/~huangrui/imagenet_ood_dataset/Places.tar.gz

3. Textures:

   • Download from the official website.

After downloading, please extract all datasets and place them in the ./dataset/ directory.

2. Generating Pseudo Masks

In this step, we use a pre-trained classification model to generate pseudo masks. These masks will be used in the subsequent training process. Use the following command to generate pseudo masks:

python make_cam.py --dataset=../dataset/cifar10/train --model_path=cls_pretrained_models/cifar10.pth.tar --head_size=10
python make_cam.py --dataset=../dataset/cifar100/train --model_path=cls_pretrained_models/cifar100.pth.tar --head_size=100

• --dataset: Specifies the path to the training dataset
• --model_path: Specifies the path to the pre-trained model
• --head_size: Sets the size of the classification head (number of classes)

3. Training

To train the model, use the following command:

bash scripts/train.sh

This script will initiate the training process using the prepared datasets and generated pseudo masks.

4. Evaluation

After training, you can evaluate the model's performance using the following command:

bash scripts/eval.sh

This script will run the evaluation process on the test set and output the results.

Note: Make sure you have completed all previous steps (dataset preparation, pseudo mask generation, and training) before running the evaluation script.

Citation

If you use our codebase, please cite our work:

@inproceedings{
ding2024improving,
title={Improving Out-of-Distribution Detection with Disentangled Foreground and Background Features},
author={Choubo Ding and Guansong Pang},
booktitle={ACM Multimedia 2024},
year={2024}
}