Research map / Label-efficient and robust learning / Recognition and long-tail
Balanced Meta-Softmax for Long-Tailed Visual 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 Balanced Softmax and Meta Sampler together (ablation) | CIFAR-10-LT and CIFAR-100-LT · CIFAR-LT, imbalance 200 | top-1 accuracy: Balanced Softmax+DT: 78.6 ± 0.1; BALMS: 81.5 ± 0.0 | Helped |
| Use Balanced Softmax instead of standard Softmax (ablation) | CIFAR-10-LT and CIFAR-100-LT · CIFAR-LT, imbalance 200 | top-1 accuracy: Softmax: 71.2 ± 0.3; Balanced Softmax: 79.0 ± 0.8 | Helped |
| Use the derived Balanced Softmax 1/4 variant instead of the full Balanced Softmax (ablation) | CIFAR-10-LT and CIFAR-100-LT · CIFAR-LT, imbalance 200 | top-1 accuracy: Balanced Softmax 1/4: 71.6 ± 0.7; Balanced Softmax: 79.0 ± 0.8 | Hurt |
| Add class-balanced sampling to Balanced Softmax with decoupled training (ablation) | Balanced Softmax · CIFAR-100-LT, imbalance 200 | top-1 accuracy: Balanced Softmax+DT: 45.1 ± 0.0; Balanced Softmax+CBS+DT: 42.0 ± 0.0 | Hurt |
| Use Meta Sampler rather than Meta Reweighter (ablation) | Balanced Softmax with decoupled training · CIFAR-LT, imbalance 200 | top-1 accuracy: CIFAR-10-LT: 81.5 ± 0.0 vs 79.2 ± 0.0; CIFAR-100-LT: 45.5 ± 0.1 vs 45.3 ± 0.3 | Mixed |
| Add decoupled training to Balanced Softmax (ablation) | Balanced Softmax · CIFAR-LT, imbalance 200 | top-1 accuracy: Balanced Softmax: 79.0 ± 0.8; Balanced Softmax+DT: 78.6 ± 0.1 | Mixed |
| Train with Meta Sampler instead of standard Softmax sampling (ablation) | Balanced Softmax · CIFAR-10-LT, imbalance 200 | top-1 accuracy: DT+MS: 73.6 ± 0.2; BALMS: 81.5 ± 0.0 | Helped |
| Replace Balanced Softmax with standard Softmax (ablation) | long-tailed classification · CIFAR-LT, imbalance 200 | top-1 accuracy: Softmax: 71.2 ± 0.3; Balanced Softmax: 79.0 ± 0.8 | Hurt |
Closest papers by meaning
- Wrapped Cauchy Distributed Angular Softmax for Long-Tailed Visual Recognition (ICML 2023 · similarity 0.91)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.87)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.87)
- Long-tailed Recognition with Model Rebalancing (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.