Research map / Label-efficient and robust learning / Recognition and long-tail
How Re-sampling Helps for Long-Tail Learning?
Jiang-Xin Shi, Tong Wei, Yuke Xiang, Yu-Feng Li
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 |
|---|---|---|---|
| class-balanced re-sampling from scratch (vs uniform sampling) on semantically clean data | LeNet · MNIST-LT | accuracy (all): 65.8 -> 90.8 | Improved |
| class-balanced re-sampling from scratch (vs uniform sampling) on data with complex contexts | ResNet-32 · CIFAR100-LT | accuracy (all): 39.1 -> 34.1 | Hurt |
| add context-shift augmentation with context bank to class-balanced re-sampling | CSA · CIFAR100-LT IR 100 | accuracy: 34.1 -> 45.8 | Improved |
| add context bank Q of head-class backgrounds (ablation) | CSA · CIFAR100-LT IR 100 | accuracy: 41.2 -> 45.8 | Improved |
| use Grad-CAM instead of CAM for background extraction (ablation) | CSA · CIFAR100-LT IR 100 | accuracy: 45.1 -> 45.8 | Improved |
| use re-sampling module only at inference instead of ensembling with uniform module (ablation) | CSA · CIFAR100-LT IR 100 | accuracy: 43.0 -> 45.8 | Improved |
| vary selection threshold delta (0 to 0.9) (ablation) | CSA · CIFAR100-LT IR 100 | accuracy: 44.83 -> 45.83 | No clear effect |
| increase context bank volume from 1x to 64x batch size (ablation) | CSA · CIFAR100-LT IR 100 | accuracy: 45.83 -> 45.26 | Hurt |
| add CSA instead of CB-RS when fine-tuning SimSiam+rwSAM pretrained model | SimSiam+rwSAM · CIFAR10-LT | accuracy: 72.8 -> 75.5 | Improved |
| add mixup on top of CSA | CSA · CIFAR100-LT IR 100 | accuracy: 45.8 -> 46.6 | Improved |
| CSA vs CMO on ImageNet-LT few-shot | ResNet-50 · ImageNet-LT | few-shot accuracy: 20.5 -> 24.1 | Improved |
Closest papers by meaning
- Rethinking Classifier Re-Training in Long-Tailed Recognition: Label Over-Smooth Can Balance (ICLR 2025 · similarity 0.88)
- Rethinking Long-tailed Dataset Distillation: A Uni-Level Framework with Unbiased Recovery and Relabeling (AAAI 2026 · similarity 0.88)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.87)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.86)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.86)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.85)
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.
