Research map / Label-efficient and robust learning / Recognition and long-tail
Risk-Bounded Distribution Reconstruction: Stable Statistic Calibration for Long-Tailed Recognition
Guanliang Liu, Wenchao Chen, Long Tian, Xuefei Cao, Hongwei Liu, Bo Chen
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 RBDR calibration instead of plain cross-entropy training | two-stage classifier · CIFAR100-LT (imbalance 200) | Top-1 accuracy: 34.80 -> 49.02 | Improved |
| use RBDR instead of FUR | feature-space reconstruction · CIFAR100-LT (imbalance 200, 100) | Top-1 accuracy: +2.82% at rho=200, +2.97% at rho=100 | Improved |
| use RBDR instead of FUR | feature-space reconstruction · ImageNet-LT | Top-1 accuracy (All): 55.50 -> 55.91 | Improved |
| use RBDR instead of FeatRecon | two-stage classifier · ImageNet-LT | Top-1 accuracy (All): 57.80 -> 55.91 | Hurt |
| replace Gaussian covariance calibrator with class-conditional latent diffusion (RBDR-DDIM) (ablation) | RBDR · ImageNet-LT and iNaturalist 2018 | Top-1 accuracy (All): 55.91 -> 55.32 (ImageNet-LT), 72.92 -> 72.35 (iNaturalist) | Hurt |
| add supportive projection to OT initialization (ablation) | RBDR · CIFAR100-LT (imbalance 100) | Top-1 accuracy (All): 52.46 -> 53.42 | Improved |
| add low-rank shrinkage on top of supportive projection (ablation) | RBDR · CIFAR100-LT (imbalance 100) | Top-1 accuracy (All): 53.42 -> 53.87 | Improved |
| add supportive projection to OT initialization (ablation) | RBDR · ImageNet-LT | Top-1 accuracy (All): 53.23 -> 55.08 | Improved |
| add low-rank shrinkage on top of supportive projection (ablation) | RBDR · ImageNet-LT | Top-1 accuracy (All): 55.08 -> 55.91 | Improved |
| add supportive projection to OT initialization (ablation) | RBDR · iNaturalist 2018 | Top-1 accuracy (All): 70.17 -> 72.28 | Improved |
| add low-rank shrinkage on top of supportive projection (ablation) | RBDR · iNaturalist 2018 | Top-1 accuracy (All): 72.28 -> 72.92 | Improved |
| evaluate RBDR vs OT initialization on the 10 scarcest CIFAR100-LT classes | RBDR · CIFAR100-LT (imbalance 100), 10 scarcest classes | per-class accuracy: turtle 5.4% -> 14.6%; +4.7% on average | Improved |
Closest papers by meaning
- Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective (NeurIPS 2021 · similarity 0.87)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.87)
- 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)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.86)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 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.
