Research map / Label-efficient and robust learning / Recognition and long-tail
Class-Conditional Sharpness-Aware Minimization for Deep Long-Tailed Recognition
Read the paperCheck 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 |
|---|---|---|---|
| use class-conditional parameter perturbation and two-stage decoupled training | ResNet-32 · CIFAR-100-LT, imbalance ratio 100 | Top-1 accuracy: 48.71 -> 50.83 | Helped |
| use class-conditional parameter perturbation and two-stage decoupled training | ResNet-32 · CIFAR-10-LT, imbalance ratio 100 | Top-1 accuracy: 82.68 -> 83.92 | Helped |
| use class-conditional parameter perturbation and two-stage decoupled training | ResNet-152 · Places-LT | Overall Top-1 accuracy: 40.6 -> 40.6 | No clear effect |
| use class-conditional parameter perturbation and two-stage decoupled training | ResNet-50 · iNaturalist 2018 | Overall Top-1 accuracy: 72.0 -> 70.9 | Hurt |
| add Stage 1 perturbation in the recommended class-specific gradient direction (ablation) | cRT on Places-LT · Places-LT | Accuracy: 37.8 -> 38.9 | Helped |
| add the recommended perturbation magnitude in a random direction (ablation) | cRT on Places-LT · Places-LT | Accuracy: 37.8 -> 37.5 | Hurt |
| add both the recommended perturbation direction and magnitude (ablation) | cRT on Places-LT · Places-LT | Accuracy: 37.8 -> 40.1 | Helped |
| add Stage 2 adversarial-feature classifier training after Stage 1 (ablation) | Stage 1 CC-SAM on Places-LT · Places-LT | Accuracy: 40.1 -> 40.6 | Helped |
| integrate existing flattening operations into long-tailed learning baselines | DLTR baselines · CIFAR-100-LT, imbalance ratio 100 | Accuracy gain | No clear effect |
Closest papers by meaning
- Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification (ICML 2025 · similarity 0.89)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.89)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.89)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.87)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.87)
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.