CVPR Findings2026
CurrMix: Curriculum-Enhanced MixUp for Long-Tailed Visual Recognition
Zhongquan Jian, Yanhao Chen, Bingbing Hu, Wenhan Lv, Shaopan Wang, Jipeng Wu, Junfeng Yao, Yang Lu, Qingqiang Wu
Abstract
Long-tailed Visual Recognition (LTVR) remains challenging due to the inherent data imbalance that biases deep learning models toward majority classes and leads to underperformance on minority classes. MixUp-based approaches alleviate this issue by synthesizing mixed images to augment training data and reduce bias, yet they typically mix only two images. Increasing the number of images in each mix could help, as it greatly boosts the representation of minority classes, but it also significantly raises optimization difficulty and slows convergence. To address this, we propose CurrMix, a novel MixUp-based method that integrates curriculum strategies to address convergence issues and fully exploit MixUp's potential for LTVR. In CurrMix, an Alternating Curriculum Sampling (ACS) strategy is introduced that alternately selects easier and harder images for mixing, dynamically adjusting convergence speed to accelerate when slow and moderate when fast. Comprehensive experiments on long-tailed datasets demonstrate Cur-rMix's superiority, with notable performance improvements over existing methods.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Long-tail learning via logit adjustment cites Decoupling Representation and Classifier for Long-Tailed Recognition
- Balanced Meta-Softmax for Long-Tailed Visual Recognition cites Decoupling Representation and Classifier for Long-Tailed Recognition
Builds on40
- CutMix: Regularization Strategy to Train Strong Classifiers With Localizable FeaturesSangdoo Yun, Dongyoon Han, Sanghyuk Chun, Seong Joon Oh et al.ICCV 2019 · 5,843 citations
- AugMix: A Simple Data Processing Method to Improve Robustness and UncertaintyDan Hendrycks, Norman Mu, Ekin Dogus Cubuk, Barret Zoph et al.ICLR 2020 · 1,572 citations
- Decoupling Representation and Classifier for Long-Tailed RecognitionBingyi Kang, Saining Xie, Marcus Rohrbach, Zhicheng Yan et al.ICLR 2020 · 1,496 citations
- Long-tail learning via logit adjustmentAditya Krishna Menon, Sadeep Jayasumana, Ankit Singh Rawat, Himanshu Jain et al.ICLR 2021 · 937 citations
- Balanced Meta-Softmax for Long-Tailed Visual RecognitionJiawei Ren, Cunjun Yu, Shunan Sheng, Xiao Ma et al.NeurIPS 2020 · 861 citations
Related papers
- CUDA: Curriculum of Data Augmentation for Long-tailed RecognitionSumyeong Ahn, Jongwoo Ko, Se-Young YunICLR 2023 · 15 citations
- Confusion-Aware Spectral Regularizer for Long-Tailed RecognitionZiquan Zhu, Gaojie Jin, Hanruo Zhu, Si-Yuan Lu et al.CVPR 2026 · 4 citations
- ConMix: Contrastive Mixup at Representation Level for Long-tailed Deep ClusteringZhixin Li, Yuheng JiaICLR 2025
- Enhancing Minority Classes by Mixing: An Adaptative Optimal Transport Approach for Long-tailed ClassificationJintong Gao, He Zhao, Zhuo Li, Dandan GuoNeurIPS 2023 · 64 citations
- Bag of Tricks for Long-Tailed Visual Recognition with Deep Convolutional Neural NetworksYongshun Zhang, Xiu-Shen Wei, Boyan Zhou, Jianxin WuAAAI 2021 · 162 citations
