/LT-DS

[ECCV 2022] Tackling Long-Tailed Category Distribution Under Domain Shifts

Primary LanguagePythonMIT LicenseMIT

LT-DS

[ECCV 2022] Repo for our paper "Tackling Long-Tailed Category Distribution Under Domain Shifts"

[project] [dataset] [paper]

figure1

Abstract

Machine learning models fail to perform well on real-world applications when

  1. LT: the category distribution P(Y) of the training dataset suffers from long-tailed distribution;
  2. DS: the test data is drawn from different conditional distributions P(X|Y).

Existing approaches cannot handle the scenario where both issues exist, which however is common for real-world applications. In this study, we took a step forward and looked into the problem of long-tailed classification under domain shifts. By taking both the categorical distribution bias and conditional distribution shifts into account, we designed three novel core functional blocks including Distribution Calibrated Classification Loss, Visual-Semantic Mapping and Semantic-Similarity Guided Augmentation. Furthermore, we adopted a meta-learning framework which integrates the three blocks to improve domain generalization on unseen target domains.

Dataset

We provide two datasets for benchmarking LT-DS (Long-Tailed Under Domain shifts) algorithms. Due to the license issue, we only provided instructions on how to create the corresponding datasets. Please follow here.

Install Env

conda env create -f requirement.yml

Training

AWA2-LTS

python train/trainer.py --cfg config/exp/awa2.yaml DATA.SOURCE SHUV DATA.TARGET SHUVO
python train/trainer.py --cfg config/exp/awa2.yaml DATA.SOURCE SHUV DATA.TARGET SHUVO MODE test

ImageNet-LTS

python train/trainer.py --cfg config/exp/imagenet.yaml DATA.SOURCE SHUV DATA.TARGET SHUVO 
python train/trainer.py --cfg config/exp/imagenet.yaml DATA.SOURCE SHUV DATA.TARGET SHUVO MODE test

TODO

  • Add requirements
  • Add evaluation scripts
  • Add imbalanced baselines

Citation

If you find our paper/code useful, please consider citing:

@inproceedings{gu2022tackling,
  title={Tackling Long-Tailed Category Distribution Under Domain Shifts},
  author={Gu, Xiao and Guo, Yao and Li, Zeju and Qiu, Jianing and Dou, Qi and Liu, Yuxuan and Lo, Benny and Yang, Guang-Zhong},
  booktitle={ECCV},
  year={2022}
  }

Acknowledgement

Our codes are inspired by the following repos: [OpenDG-DAML] [BagofTricks-LT ] [ISDA].