Fast Decision Boundary based Out-of-Distribution Detector

This repository contains code for the paper Fast Decision Boundary based Out-of-Distribution Detector (ICML 2024) by Litian Liu and Yao Qin.

Besides this codebase, our algorithm fDBD is also available on the OpenOOD Benchmark.

Setup

# create conda env and install dependencies
$ conda env create -f environment.yml
$ conda activate fdbd
# set environmental variables
$ export DATASETS='data'
# download datasets and checkpoints
$ bash download.sh

Please download ImageNet dataset manually to your own $IMAGENET dir by following this instructions.

Demo

CIFAR-10 Benchmark

python feat_extract.py --in-dataset CIFAR-10  --out-datasets SVHN iSUN dtd places365 --name resnet18-supcon  --model-arch resnet18-supcon

python run_cifar.py --in-dataset CIFAR-10  --out-datasets SVHN iSUN dtd places365 --name resnet18-supcon   --model-arch resnet18-supcon 

ImageNet Benchmark

python feat_extract_largescale.py --in-dataset imagenet --imagenet-root $IMAGENET --out-datasets inat sun50 places50 dtd  --name resnet50-supcon --model-arch resnet50-supcon

python run_imagenet.py --in-dataset imagenet  --out-datasets inat sun50 places50 dtd  --name resnet50-supcon  --model-arch resnet50-supcon

fDBD with activation shaping

python feat_extract_largescale.py --in-dataset imagenet --imagenet-root /data/chenhe/datasets/imagenet --out-datasets inat sun50 places50 dtd  --name resnet50 --model-arch resnet50

python run_imagenet_w_ASH.py --in-dataset imagenet  --out-datasets inat sun50 places50 dtd  --name resnet50  --model-arch resnet50

Inference Efficiency

python e2e_timing.py --in-dataset CIFAR-10 --name resnet18 --model-arch resnet18

References

@article{djurisic2022ash,
    url = {https://arxiv.org/abs/2209.09858},
    author = {Djurisic, Andrija and Bozanic, Nebojsa and Ashok, Arjun and Liu, Rosanne},
    title = {Extremely Simple Activation Shaping for Out-of-Distribution Detection},
    publisher = {arXiv},
    year = {2022},
    }

@article{sun2022knnood,
  title={Out-of-distribution Detection with Deep Nearest Neighbors},
  author={Sun, Yiyou and Ming, Yifei and Zhu, Xiaojin and Li, Yixuan},
  journal={ICML},
  year={2022}
}

Citations

Please cite our paper if you find this codebase helpful!

@article{liu2024fast,
  title={Fast Decision Boundary based Out-of-Distribution Detector},
  author={Liu, Litian and Qin, Yao},
  journal={ICML},
  year={2024}
}