Research map / Label-efficient and robust learning / Recognition and long-tail
Decoupled and calibrated long-tail training
Rebalances optimization and calibrates statistics between head and tail classes.
82 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in Recognition and long-tail, keyword cluster “long tailed · classes · tailed recognition”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
25 papers in 2023–24 and 26 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 2.0% to 2.0%.
No longer reported by papers here: iNaturalist 2018 (top-1 accuracy, last in 2022); ImageNet-LT (top-1 accuracy (%), last in 2022); CIFAR-100-LT IF 100 (top-1 accuracy, last in 2023); CIFAR100-LT, imbalance ratio 100 (top-1 accuracy, last in 2024); CIFAR100-LT, imbalance ratio 50 (top-1 accuracy, last in 2024); ImageNet-LT (accuracy, last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Large-Scale Long-Tailed Recognition in an Open World (CVPR 2019 · 1,159 citations)
- 2020 · BBN: Bilateral-Branch Network With Cumulative Learning for Long-Tailed Visual Recognition (CVPR 2020 · 905 citations)
- 2021 · Contrastive Learning Based Hybrid Networks for Long-Tailed Image Classification (CVPR 2021 · 298 citations)
- 2022 · Balanced Contrastive Learning for Long-Tailed Visual Recognition (CVPR 2022 · 221 citations)
- 2023 · Global and Local Mixture Consistency Cumulative Learning for Long-Tailed Visual Recognitions (CVPR 2023 · 99 citations)
- 2024 · Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · 52 citations)
- 2025 · MLC-NC: Long-Tailed Multi-Label Image Classification Through the Lens of Neural Collapse (AAAI 2025 · 4 citations)
- 2026 · BCE3S: Binary Cross-Entropy Based Tripartite Synergistic Learning for Long-Tailed Recognition (AAAI 2026 · 1 citation)
Earliest papers here
- Large-Scale Long-Tailed Recognition in an Open World (CVPR 2019 · 1,159 citations)
- BBN: Bilateral-Branch Network With Cumulative Learning for Long-Tailed Visual Recognition (CVPR 2020 · 905 citations)
- Balanced Meta-Softmax for Long-Tailed Visual Recognition (NeurIPS 2020 · 315 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
iNaturalist 2018 · top-1 accuracy (higher is better)
6 papers, 2021–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 71.6 | Improving Calibration for Long-Tailed Recognition (CVPR 2021 · 281 citations) |
| 2022 | 72.01 | Long-Tailed Visual Recognition via Gaussian Clouded Logit Adjustment (CVPR 2022 · 92 citations) |
ImageNet-LT · top-1 accuracy (%) (higher is better)
5 papers, 2021–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 55.4 | Long-tailed Recognition by Routing Diverse Distribution-Aware Experts (ICLR 2021 · 30 citations) |
| 2022 | 56 | Balanced Contrastive Learning for Long-Tailed Visual Recognition (CVPR 2022 · 221 citations) |
CIFAR-100-LT IF 100 · top-1 accuracy (higher is better)
4 papers, 2021–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 47 | Improving Calibration for Long-Tailed Recognition (CVPR 2021 · 281 citations) |
| 2022 | 54.2 | Nested Collaborative Learning for Long-Tailed Visual Recognition (CVPR 2022 · 106 citations) |
| 2023 | 55.88 | Global and Local Mixture Consistency Cumulative Learning for Long-Tailed Visual Recognitions (CVPR 2023 · 99 citations) |
CIFAR100-LT, imbalance ratio 100 · top-1 accuracy (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 45.4 | Disentangling Label Distribution for Long-Tailed Visual Recognition (CVPR 2021 · 255 citations) |
| 2022 | 48.71 | Long-Tailed Visual Recognition via Gaussian Clouded Logit Adjustment (CVPR 2022 · 92 citations) |
| 2023 | 50.83 | Class-Conditional Sharpness-Aware Minimization for Deep Long-Tailed Recognition (CVPR 2023 · 46 citations) |
| 2024 | 51.38 | Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · 52 citations) |
CIFAR100-LT, imbalance ratio 50 · top-1 accuracy (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 50.5 | Disentangling Label Distribution for Long-Tailed Visual Recognition (CVPR 2021 · 255 citations) |
| 2022 | 53.55 | Long-Tailed Visual Recognition via Gaussian Clouded Logit Adjustment (CVPR 2022 · 92 citations) |
| 2023 | 53.91 | Class-Conditional Sharpness-Aware Minimization for Deep Long-Tailed Recognition (CVPR 2023 · 46 citations) |
| 2024 | 55.54 | Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · 52 citations) |
ImageNet-LT · accuracy (higher is better)
4 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 43.2 | Inflated Episodic Memory With Region Self-Attention for Long-Tailed Visual Recognition (CVPR 2020 · 84 citations) |
| 2023 | 58.3 | Long-Tailed Recognition by Mutual Information Maximization between Latent Features and Ground-Truth Labels (ICML 2023 · 4 citations) |
What has been tried
59 papers here report 683 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 5%, hurt 20%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use instance-wise sampling rather than reverse sampling in the fusing branch | CIFAR100-LT, imbalance ratio 100 · Tail accuracy | 24.07 -> 27.07 | Helped | Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · 52 citations) |
| add IBR to Base | PASCAL VOC 2012 training set CAM · mIoU (%) | 55.1 -> 55.9 | Helped | SFC: Shared Feature Calibration in Weakly Supervised Semantic Segmentation (AAAI 2024 · 51 citations) |
| Use three stages instead of two | ImageNet-LT test · accuracy (Avg) | 50.3 -> 60.4 | Helped | Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · 12 citations) |
| Add generated samples to training | CIFAR100-LT (r=100) · accuracy | 38.3 -> 46.6 | Helped | DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge (NeurIPS 2024 · 8 citations) |
| add ETF label embedding guided feature learning | VOC-LT · mAP (%) — Total | 82.13 -> 84.37 | Helped | MLC-NC: Long-Tailed Multi-Label Image Classification Through the Lens of Neural Collapse (AAAI 2025 · 4 citations) |
| add DRW to VPT training | CIFAR100-LT (IR 100) · Accuracy | 88.25 -> 89.32 | Helped | Improving Visual Prompt Tuning by Gaussian Neighborhood Minimization for Long-Tailed Visual Recognition (NeurIPS 2024 · 4 citations) |
| Add weight decay (1e-04) to curvature scoring | CIFAR100 · cosine similarity with FZ scores | 0.73 -> 0.82 | Helped | Memorization Through the Lens of Curvature of Loss Function Around Samples (ICML 2024 · 1 citation) |
| Increase gamma from a moderate value to a larger value | CIFAR-LT · accuracy | Mixed | Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification (ICML 2025 · 0 citations) | |
| Calibrate labels from an imbalanced labeling model | CIFAR-10-LT, IF=10, IPC=10 · Accuracy | 34.3 -> 41.4 | Helped | Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation (NeurIPS 2025 · 0 citations) |
| remove logit clipping during estimator training | Backward 50 · accuracy | 57.3 -> 48.2 | Hurt | Learning Label Shift Correction for Test-Agnostic Long-Tailed Recognition (ICML 2024) |
| Add TFA, FLSA, and APDS | AVA · MAE (low) | 0.616 -> 0.438 | Helped | ELTA: An Enhancer against Long-Tail for Aesthetics-oriented Models (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10-LT · 5 of 12 papers
- CIFAR-100-LT · 5 of 12 papers
- ImageNet-LT · 4 of 12 papers
- iNaturalist 2018 · 2 of 12 papers
Baselines
- Focal Loss · 2 of 12 papers
- BBN · 2 of 12 papers
- LDAM+DRW · 2 of 12 papers
- SBCL · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 51 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 7 papers
- Chinese Academy of Sciences · 4 papers
- University of Chinese Academy of Sciences · 4 papers
- Nanjing University · 3 papers
- Jilin University · 3 papers
- Shenzhen University · 2 papers
- Xiamen University · 2 papers
- Huazhong University of Science and Technology · 2 papers
- Tencent · 2 papers
- Shanghai Jiao Tong University · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Classifier rebalancing and margin losses (58 papers)
- Contrastive learning for imbalanced data (22 papers)
Elsewhere in the topic
- Multilingual language representation pretraining (507 papers)
- Contrastive representation theory (495 papers)
- Dense features and foundation priors (482 papers)
- Correspondence and causal representation learning (447 papers)
- Data augmentation and generalization bias (409 papers)
- Partial and multi-label learning (345 papers)
- Loss design and classifier theory (273 papers)
- Group-robust classification (251 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).