Research map / Label-efficient and robust learning / Recognition and long-tail
Long-Tailed Visual Recognition via Gaussian Clouded Logit Adjustment
Check 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 Gaussian clouded logits with class-dependent cloud sizes | GCL · CIFAR-10-LT, imbalance ratio 200 | top-1 accuracy (%): 79.03 vs 77.31 | Helped |
| Use Gaussian clouded logits with class-dependent cloud sizes | GCL · CIFAR-100-LT, imbalance ratio 200 | top-1 accuracy (%): 44.88 vs 42.33 | Helped |
| Use Gaussian clouded logits with class-dependent cloud sizes | GCL · ImageNet-LT | top-1 accuracy (%): 54.88 vs 52.91 | Helped |
| Use Gaussian clouded logits with class-dependent cloud sizes | GCL · iNaturalist 2018 | top-1 accuracy (%): 72.01 vs 71.57 | Helped |
| Use Gaussian clouded logits with class-dependent cloud sizes | GCL · Places-LT | top-1 accuracy (%): 40.64 vs 40.15 | Helped |
| Use logarithmic-difference cloud sizing rather than cosine, power-difference e:1/3, or power-difference e:1/4 (ablation) | GCL · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 79.21; 80.80; 82.31 -> 82.68 | Helped |
| Use CBEN rather than instance-balance sampling (ablation) | GCL classifier re-balancing · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 80.41 -> 82.68 | Helped |
| Use CBEN rather than class-balance sampling (ablation) | GCL classifier re-balancing · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 82.43 -> 82.68 | Helped |
| Use CBEN rather than effective-number sampling (ablation) | GCL classifier re-balancing · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 82.47 -> 82.68 | Helped |
| Add classifier re-training (ablation) | GCL · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 80.52 -> 82.68 | Helped |
| Use cRT rather than learnable weight scaling (ablation) | GCL classifier re-balancing · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 82.25 -> 82.68 | Helped |
| Use cRT rather than tau-normalization (ablation) | GCL classifier re-balancing · CIFAR-10-LT, imbalance ratio 100 | top-1 accuracy (%): 82.16 -> 82.68 | Helped |
Closest papers by meaning
- Balanced Meta-Softmax for Long-Tailed Visual Recognition (NeurIPS 2020 · similarity 0.88)
- Exploring Weight Balancing on Long-Tailed Recognition Problem (ICLR 2024 · similarity 0.86)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.86)
- Improving Visual Prompt Tuning by Gaussian Neighborhood Minimization for Long-Tailed Visual Recognition (NeurIPS 2024 · similarity 0.85)
- 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.