DirMixE

This repository is the official code for the paper Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition (ICML 2024).

Paper Title: Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition.

Author: Zhiyong Yang, Qianqian Xu*, Zitai Wang, Sicong Li, Boyu Han, Shilong Bao, Xiaochun Cao, Qingming Huang*

🔥 News

• [2026] Our work has been further extended in a TPAMI version with improved methodology and more comprehensive experiments.
  👉 Please check the new repository: https://github.com/scongl/DirMixE_TPAMI

Installation

• Clone the repository

git clone https://github.com/scongl/DirMixE.git

• Install the required libraries

pip install -r requirements.txt

Datasets

Four bechmark datasets

• Please download these datasets and put them to the /data file.
• ImageNet-LT can be found at here.
• iNaturalist data should be the 2018 version from here.
• CIFAR-100 and CIFAR-10 will be downloaded automatically with the dataloader.

data
├── CIFAR-10
│   └── cifar-10-batches-py
├── CIFAR-100
│   └── cifar-100-python
├── ImageNet_LT
│   ├── test
│   ├── train
│   └── val
└── iNaturalist
    ├── test2018
    └── train_val2018

Txt files

• We provide txt files for test-agnostic long-tailed recognition for ImageNet-LT and iNaturalist 2018. CIFAR-100 and CIFAR-10 will be generated automatically with the code.
• For iNaturalist 2018, please unzip the iNaturalist_train.zip.

data_txt
├── ImageNet_LT
│   ├── ImageNet_LT_backward10.txt
│   ├── ImageNet_LT_backward25.txt
│   ├── ImageNet_LT_backward2.txt
│   ├── ImageNet_LT_backward50.txt
│   ├── ImageNet_LT_backward5.txt
│   ├── ImageNet_LT_forward10.txt
│   ├── ImageNet_LT_forward25.txt
│   ├── ImageNet_LT_forward2.txt
│   ├── ImageNet_LT_forward50.txt
│   ├── ImageNet_LT_forward5.txt
│   ├── ImageNet_LT_test.txt
│   ├── ImageNet_LT_train.txt
│   ├── ImageNet_LT_uniform.txt
│   └── ImageNet_LT_val.txt
└── iNaturalist18
    ├── iNaturalist18_backward2.txt
    ├── iNaturalist18_backward3.txt
    ├── iNaturalist18_forward2.txt
    ├── iNaturalist18_forward3.txt
    ├── iNaturalist18_train.txt
    ├── iNaturalist18_uniform.txt
    └── iNaturalist18_val.txt

Script

CIFAR10-LT

Training

• To train the DirMixE model, run this command:

python train.py -c configs/train/cifar10.json 

Evaluate

• To evaluate DirMixE under ours setting, run this command:

python test_train_cifar_dirmixe.py -c configs/test/test_cifar10.json -r checkpoint_path

• To evaluate DirMixE under sade's setting, run this command:

python test_train_cifar_sade.py -c configs/test/test_cifar10.json -r checkpoint_path

CIFAR100-LT

Training

• To train the DirMixE model, run this command:

python train.py -c configs/train/cifar100.json 

Evaluate

• To evaluate DirMixE under ours setting, run this command:

python test_train_cifar_dirmixe.py -c configs/test/test_cifar100.json -r checkpoint_path

• To evaluate DirMixE under sade's setting, run this command:

python test_train_cifar_sade.py -c configs/test/test_cifar100.json -r checkpoint_path

ImageNet-LT

Training

• To train the DirMixE model, run this command:

python train.py -c configs/train/imagenet.json 

Evaluate

• To evaluate DirMixE under ours setting, run this command:

python test_train_cifar_dirmixe.py -c configs/test/test_imagenet.json -r checkpoint_path

• To evaluate DirMixE under sade's setting, run this command:

python test_train_cifar_sade.py -c configs/test/test_imagenet.json -r checkpoint_path

iNaturalist

Training

• To train the DirMixE model, run this command:

python train.py -c configs/train/inaturalist.json

Evaluate

• To evaluate DirMixE under ours setting, run this command:

python test_train_cifar_dirmixe.py -c configs/test/test_inaturalist.json -r checkpoint_path

• To evaluate DirMixE under sade's setting, run this command:

python test_train_cifar_sade.py -c configs/test/test_inaturalist.json -r checkpoint_path

Citation

If you find our work inspiring or use our codebase in your research, please cite our work.

@inproceedings{yang2024harnessing,
    title={Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition}, 
    author={Zhiyong Yang and Qianqian Xu and Zitai Wang and Sicong Li and Boyu Han and Shilong Bao and Xiaochun Cao and Qingming Huang},
    booktitle={International Conference on Machine Learning},
    year={2024}
}

@article{yang2026dirmixe,
    title = {DirMixE: Harnessing Test Agnostic Long-tail Recognition with Hierarchical Label Vartiations},
    author = {Zhiyong Yang and Qianqian Xu and Sicong Li and Zitai Wang and Xiaochun Cao and Qingming Huang},
    journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
    year = {2026},
}

Acknowledgements

Our codes are based on the repository SADE.