Research map / Label-efficient and robust learning / Recognition and long-tail
Balanced Contrastive Learning for Long-Tailed Visual Recognition
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 |
|---|---|---|---|
| replace supervised contrastive learning with BCL | ResNet-32 long-tailed CIFAR training · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 48.21 -> 51.93 | Helped |
| replace supervised contrastive learning with BCL | ResNet-32 long-tailed CIFAR training · CIFAR-10-LT, imbalance factor 10 | Top-1 accuracy: 91.12 -> 91.12 | No clear effect |
| add a supervised contrastive branch to logit-compensated cross-entropy (ablation) | CIFAR-100-LT, imbalance factor 100 · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 50.8 -> 52.4 | Helped |
| add class-complement to LC plus supervised contrastive learning (ablation) | CIFAR-100-LT, imbalance factor 100 · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 52.4 -> 52.3 | No clear effect |
| add class-averaging to LC plus supervised contrastive learning (ablation) | CIFAR-100-LT, imbalance factor 100 · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 52.4 -> 52.0 | Hurt |
| combine class-complement and class-averaging (ablation) | LC plus supervised contrastive learning · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 52.4 -> 53.9 | Helped |
| average class similarities outside rather than inside the exponential (ablation) | BCL class-averaging · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 50.2 -> 51.9 | Helped |
| use class prototypes instead of the batch embedding average (ablation) | BCL class-averaging · CIFAR-100-LT, imbalance factor 100 | Top-1 accuracy: 50.2 -> 51.0 | Helped |
| replace classifier-weight prototypes with exponential moving average prototypes (ablation) | BCL · CIFAR-100-LT, imbalance factor 100, 400 epochs | Top-1 accuracy (All): 53.9 -> 53.0 | Hurt |
| replace classifier-weight prototypes with learnable parameter prototypes (ablation) | BCL · CIFAR-100-LT, imbalance factor 100, 400 epochs | Top-1 accuracy (All): 53.9 -> 53.3 | Hurt |
| use RandAug for one view instead of SimAug for both views (ablation) | BCL contrastive branch · ImageNet-LT, 90 epochs | Top-1 accuracy (All): 56.7 -> 57.1 | Helped |
| use RandAug for both views instead of RandAug and SimAug (ablation) | BCL contrastive branch · ImageNet-LT, 90 epochs | Top-1 accuracy (All): 57.1 -> 57.3 | Helped |
Closest papers by meaning
- Decoupled Contrastive Learning for Long-Tailed Recognition (AAAI 2024 · similarity 0.93)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.91)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.90)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.89)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.89)
- BaCon: Boosting Imbalanced Semi-supervised Learning via Balanced Feature-Level Contrastive Learning (AAAI 2024 · similarity 0.88)
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.