Research map / Label-efficient and robust learning / Recognition and long-tail
The Majority Can Help the Minority: Context-Rich Minority Oversampling for Long-Tailed Classification
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 context-rich minority oversampling to RIDE (3 experts) | RIDE (3 experts) · CIFAR-100-LT, imbalance ratio 100 | accuracy: 48.6 -> 50.0 | Helped |
| Add context-rich minority oversampling to CE | CE · CIFAR-100-LT, imbalance ratio 100 | accuracy: 38.6 -> 43.9 | Helped |
| Add context-rich minority oversampling to RIDE (3 experts) | RIDE (3 experts) · ImageNet-LT | All accuracy: 54.9 -> 56.2 | Helped |
| Add context-rich minority oversampling to CE | CE · ImageNet-LT | All accuracy: 41.6 -> 49.1 | Helped |
| Add context-rich minority oversampling to Balanced Softmax | Balanced Softmax · ImageNet-LT, 400 epochs with RandAugment | All accuracy: 55.0 -> 58.0 | Helped |
| Add context-rich minority oversampling to CE | CE · iNaturalist 2018 | All accuracy: 61.0 -> 68.9 | Helped |
| Use q(1,k) rather than q(1/2,k) for foreground sampling (ablation) | CMO · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 42.6 -> 43.9 | Helped |
| Use q(1,k) rather than q(2,k) for foreground sampling (ablation) | CMO · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 40.1 -> 43.9 | Helped |
| Use q(1,k) rather than effective-number sampling for foreground sampling (ablation) | CMO · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 39.5 -> 43.9 | Helped |
| Use original-distribution backgrounds and minor-class-weighted foregrounds instead of sampling both from the weighted distribution (ablation) | CMO with CE · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 37.9 -> 43.9 | Helped |
| Use original-distribution backgrounds and minor-class-weighted foregrounds instead of weighted backgrounds and original-distribution foregrounds (ablation) | CMO with CE · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 40.1 -> 43.9 | Helped |
| Use minor-class-weighted foreground sampling rather than unweighted CutMix sampling (ablation) | CutMix with CE · CIFAR-100-LT, imbalance ratio 100 | All accuracy: 35.6 -> 43.9 | Helped |
Closest papers by meaning
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.86)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.86)
- Concentration and excess risk bounds for imbalanced classification with synthetic oversampling (NeurIPS 2025 · similarity 0.85)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.85)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.85)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (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.