Out-of-Distribution Detection in Long-Tailed Recognition with Calibrated Outlier Class Learning

This is the official implementation of the AAAI'24 paper titled Out-of-Distribution Detection in Long-Tailed Recognition with Calibrated Outlier Class Learning

Dataset Preparation

In-distribution dataset

Please download CIFAR10, CIFAR100, and ImageNet-LT , place them in./datasets

Auxiliary/Out-of-distribution dataset

For CIFAR10-LT and CIFAR100-LT, please download TinyImages 300K Random Images for auxiliary in ./datasets

For CIFAR10-LT and CIFAR100-LT, please download SC-OOD benchmark for out-of-distribution in ./datasets

For ImageNet-LT, please download ImageNet10k_eccv2010 benchmark for auxiliary and out-of-distribution in ./datasets

All datasets follow PASCL

Training

CIFAR10-LT:

python train.py --gpu 0 --ds cifar10 --Lambda1 0.05 --Lambda2 0.05 --Lambda3 0.1 --drp <where_you_store_all_your_datasets> --srp <where_to_save_the_ckpt>

CIFAR100-LT:

python train.py --gpu 0 --ds cifar100 --Lambda1 0.05 --Lambda2 0.05 --Lambda3 0.1  --drp <where_you_store_all_your_datasets> --srp <where_to_save_the_ckpt>

ImageNet-LT:

python stage1.py --gpu 0,1,2,3 --ds imagenet --md ResNet50 --lr 0.1 --Lambda1 0.02 --Lambda2 0.01 --Lambda3 0.01 --drp <where_you_store_all_your_datasets> --srp <where_to_save_the_ckpt>

Testing

CIFAR10-LT:

for dout in texture svhn cifar tin lsun places365
do
python test.py --gpu 0 --ds cifar10 --dout $dout \
    --drp <where_you_store_all_your_datasets> \
    --ckpt_path <where_you_save_the_ckpt>
done

CIFAR100-LT:

for dout in texture svhn cifar tin lsun places365
do
python test.py --gpu 0 --ds cifar100 --dout $dout \
    --drp <where_you_store_all_your_datasets> \
    --ckpt_path <where_you_save_the_ckpt>
done

ImageNet-LT:

python test_imagenet.py --gpu 0 \
    --drp <where_you_store_all_your_datasets> \
    --ckpt_path <where_you_save_the_ckpt>

Acknowledgment

Part of our codes are adapted from these repos:

Outlier-Exposure - https://github.com/hendrycks/outlier-exposure - Apache-2.0 license

PASCL - https://github.com/amazon-science/long-tailed-ood-detection - Apache-2.0 license

Open-Sampling - https://github.com/hongxin001/open-sampling - Apache-2.0 license

Long-Tailed-Recognition.pytorch - https://github.com/KaihuaTang/Long-Tailed-Recognition.pytorch - GPL-3.0 license

License

This project is licensed under the Apache-2.0 License.

Citation

If you use this package and find it useful, please cite our paper using the following BibTeX.

@inproceedings{miao2024out,
  title={Out-of-distribution detection in long-tailed recognition with calibrated outlier class learning},
  author={Miao, Wenjun and Pang, Guansong and Bai, Xiao and Li, Tianqi and Zheng, Jin},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={5},
  pages={4216--4224},
  year={2024}
}