Research map / Label-efficient and robust learning / Recognition and long-tail
ImbSAM: A Closer Look at Sharpness-Aware Minimization in Class-Imbalanced Recognition
Yixuan Zhou, Yi Qu, Xing Xu, Hengtao Shen
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 |
|---|---|---|---|
| apply class-aware ImbSAM instead of class-agnostic SAM | CE · CIFAR100-LT (IF=100) | Few accuracy (%): 12.4 -> 19.4 | Improved |
| apply class-aware ImbSAM instead of class-agnostic SAM | LTR · CIFAR100-LT (IF=100) | Few accuracy (%): 25.3 -> 39.4 | Improved |
| add standard SAM to CE | CE · CIFAR100-LT (IF=100) | Few accuracy (%): 13.5 -> 12.4 | Hurt |
| add large tail loss re-weighting (up to 20) to SAM (ablation) | CE+SAM · CIFAR100-LT (IF=100) | All accuracy (%): 48.9 -> 49.1 | No clear effect |
| apply ImbSAM instead of CE alone | CE · CIFAR100-LT (IF=100) | Many accuracy (%): 77.8 -> 75.9 | Hurt |
| apply ImbSAM on top of WB (LTR) | LTR · CIFAR100-LT (IF=100) | All accuracy (%): 51.9 -> 54.8 | Improved |
| set class split threshold eta to the Few-class limit (20) (ablation) | ImbSAM · CIFAR100-LT (IF=100) | Few accuracy (%): up to 21.5% | Improved |
| increase rho with eta=20 (ablation) | ImbSAM · CIFAR100-LT (IF=100) | Few accuracy (%) | Hurt |
| apply standard SAM on large-scale long-tailed data | CE · ImageNet-LT and iNaturalist | accuracy on Many classes | No clear effect |
| apply ImbSAM to DeepSAD | DeepSAD · SSAD average of 5 datasets | AUCROC (%): 90.6 -> 91.8 | Improved |
| apply standard SAM to DeepSAD and DevNet | DeepSAD, DevNet · SSAD average of 5 datasets | AUCROC (%): 90.6 -> 90.4; 91.6 -> 91.0 | Hurt |
| apply ImbSAM to DevNet at low anomaly ratio | DevNet · SSAD AUCPR anomaly (gamma_l<0.1) | AUCPR anomaly | Hurt |
Closest papers by meaning
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
- Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification (ICML 2025 · similarity 0.88)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.88)
- SSE-SAM: Balancing Head and Tail Classes Gradually Through Stage-Wise SAM (AAAI 2025 · similarity 0.87)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.86)
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.
