Research map / Label-efficient and robust learning / Recognition and long-tail
Class Rectification Hard Mining for Imbalanced Deep Learning
Qi Dong, Shaogang Gong, Xiatian Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CRL with instance-level hard mining | DeepID2 · CelebA | average class-balanced accuracy: 81 -> 86 | Improved |
| use CRL(I) instead of LMLE | imbalanced learning on DeepID2 · CelebA | average class-balanced accuracy: 84 -> 86 | Improved |
| use CRL(I) instead of LMLE on the 20 most imbalanced attributes | imbalanced learning · CelebA | average accuracy: 2% -> 7% | Improved |
| use CRL(I) instead of LMLE | imbalanced learning · X-Domain | average class-balanced accuracy: 75.77 -> 79.66 | Improved |
| add class-level mining CRL(C) instead of LMLE | imbalanced learning · X-Domain | average class-balanced accuracy: 75.77 -> 78.24 | Improved |
| add over-sampling | DeepID2 · CelebA | average class-balanced accuracy: 81 -> 82 | No clear effect |
| add down-sampling | DeepID2 · CelebA | average class-balanced accuracy: 81 -> 78 | Hurt |
| add down-sampling | MTCT · X-Domain | average class-balanced accuracy: 73.53 -> 26.29 | Hurt |
| use instance-level instead of class-level hard mining with relative loss (ablation) | CRL · CelebA | additional gain in average accuracy: 4.23 -> 5.85 | Improved |
| use instance-level instead of class-level hard mining with relative loss (ablation) | CRL · X-Domain | additional gain in average accuracy: 4.71 -> 6.13 | Improved |
| use instance-level instead of class-level hard mining with absolute loss (ablation) | CRL · CelebA | additional gain in average accuracy: 5.71 -> 5.67 | No clear effect |
| use distribution comparison loss instead of relative comparison loss (instance-level) (ablation) | CRL · CelebA | additional gain in average accuracy: 5.85 -> 2.12 | Hurt |
Closest papers by meaning
- Imbalanced Deep Learning by Minority Class Incremental Rectification (TPAMI 2018 · similarity 0.93)
- Deep Imbalanced Learning for Face Recognition and Attribute Prediction (TPAMI 2019 · similarity 0.89)
- A Unified Loss for Handling Inter-Class and Intra-Class Imbalance in Medical Image Segmentation (AAAI 2025 · similarity 0.85)
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · similarity 0.85)
- Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance (ICML 2024 · similarity 0.85)
- Synthetic Tabular Data Generation for Imbalanced Classification: The Surprising Effectiveness of an Overlap Class (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
