Research map / Label-efficient and robust learning / Recognition and long-tail
No One Left Behind: Improving the Worst Categories in Long-Tailed Learning
Yingxiao Du, Jianxin Wu
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 |
|---|---|---|---|
| Fine-tune PaCo's classifier with GML | PaCo · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 36.42 -> 39.20 | Improved |
| Fine-tune PaCo's classifier with GML | PaCo · CIFAR100-LT (imbalance ratio 100) | Lowest Recall: 5.00 -> 9.00 | Improved |
| Ensemble the original and GML-fine-tuned classifiers | PaCo + GML · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 39.20 -> 41.02 | Improved |
| Ensemble the original and GML-fine-tuned classifiers | PaCo + GML · CIFAR100-LT (imbalance ratio 100) | Accuracy: 50.53 -> 49.82 | Hurt |
| Fine-tune PaCo's classifier with GML | PaCo · ImageNet-LT | Harmonic Mean: 21.79 -> 43.71 | Improved |
| Fine-tune PaCo's classifier with GML | PaCo · ImageNet-LT | Lowest Recall: 0.00 -> 2.00 | Improved |
| Fine-tune MiSLAS's classifier with GML and label-aware smoothing | MiSLAS · Places-LT | Harmonic Mean: 28.97 -> 29.98 | Improved |
| Remove re-weighting from GML (ablation) | CE + GML · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 31.26 -> 5.11 | Hurt |
| Replace GML re-weighting with re-sampling (ablation) | CE + GML · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 31.26 -> 23.05 | Hurt |
| Use cross-entropy with re-weighting instead of GML for fine-tuning (ablation) | CE fine-tuning · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 31.26 -> 27.55 | Hurt |
| Use cross-entropy with re-sampling instead of GML for fine-tuning (ablation) | CE fine-tuning · CIFAR100-LT (imbalance ratio 100) | Lowest Recall: 6.00 -> 2.00 | Hurt |
| Train from scratch with GML rather than attaching it to a pretrained model (ablation) | GML · CIFAR100-LT (imbalance ratio 100) | Harmonic Mean: 31.26 -> 30.86; 39.20 -> 30.86 | Mixed |
Closest papers by meaning
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.90)
- Exploring Weight Balancing on Long-Tailed Recognition Problem (ICLR 2024 · similarity 0.89)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.89)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.89)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.89)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.88)
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.
