Abstract
Long-tailed category distribution combined with domain shifts presents significant challenges in medical image analysis, where rare conditions may be underrepresented and domain variations can further exacerbate the imbalance. Traditional approaches often struggle with both challenges simultaneously. In this work, we propose a novel approach for tackling long-tailed category distribution under domain shifts.
Our method introduces a comprehensive framework that addresses both long-tailed distribution and domain shift challenges. The framework consists of three key components: (1) a balanced sampling strategy that addresses category imbalance, (2) a domain adaptation module that handles domain shifts, and (3) a unified training approach that optimizes for both challenges simultaneously.
We evaluate our approach on multiple medical imaging datasets with long-tailed distributions and domain shifts. Experimental results demonstrate that our method significantly improves performance on rare categories while maintaining robustness across domains. The approach shows particular effectiveness in handling the challenging combination of long-tailed distribution and domain shifts.
The proposed framework represents a significant advancement in handling complex real-world scenarios in medical imaging, providing more robust and balanced performance across different categories and domains.
BibTeX
@inproceedings{gu2022ci,
title={Tackling Long-Tailed Category Distribution Under Domain Shifts},
author={Gu, Xiao and Guo, Yao and Li, Zeju and Qiu Jianning and Dou, Qi and Liu, Yuxuan and Lo, Benny and Yang, Guang-Zhong},
year={2022},
booktitle={European Conference on Computer Vision (ECCV 2022)},
doi={10.1007/978-3-031-20050-2_42}
}