Research map / Label-efficient and robust learning / Recognition and long-tail
Contrastive Learning Based Hybrid Networks for Long-Tailed Image Classification
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 Hybrid-SC instead of the strongest listed baseline | long-tailed CIFAR-10 and CIFAR-100 · imbalance ratios 100, 50, and 10 | top-1 accuracy: 81.40 vs 80.6; 85.36 vs 83.60; 91.12 vs 88.50; 46.72 vs 44.10; 51.87 vs 50.30; 63.05 vs 59.60 | Helped |
| Use Hybrid-PSC instead of the strongest listed baseline | long-tailed CIFAR-10 and CIFAR-100 · imbalance ratios 100, 50, and 10 | top-1 accuracy: 78.82 vs 80.6; 83.86 vs 83.60; 90.06 vs 88.50; 44.97 vs 44.10; 48.93 vs 50.30; 62.37 vs 59.60 | Mixed |
| Use Hybrid-PSC instead of BBN | iNaturalist 2018 · iNaturalist 2018 | top-1 accuracy: 68.10 vs 66.29 | Helped |
| Use Hybrid-PSC instead of Hybrid-SC | iNaturalist 2018 · iNaturalist 2018 | top-1 accuracy: 66.74 -> 68.10 | Helped |
| Train Hybrid-PSC for 200 rather than the default 100 epochs | Hybrid-PSC · iNaturalist 2018 | top-1 accuracy: 68.10 -> 70.35 | Helped |
| Replace supervised contrastive feature learning with cross-entropy feature learning (ablation) | Hybrid network · Long-tailed CIFAR-100, imbalance ratios 100, 50, and 10 | top-1 accuracy: 46.72 -> 41.40; 51.87 -> 46.68; 63.05 -> 59.14 | Hurt |
| Use two-stage SC training instead of curriculum joint training (ablation) | Hybrid-SC · Long-tailed CIFAR-100, imbalance ratios 100, 50, and 10 | top-1 accuracy: 46.72 -> 42.73; 51.87 -> 46.76; 63.05 -> 60.62 | Hurt |
| Replace the curriculum with fixed equal loss weights, alpha=0.5 (ablation) | Hybrid-SC · Long-tailed CIFAR-100, imbalance ratios 100, 50, and 10 | top-1 accuracy: 46.72 -> 42.58; 51.87 -> 47.45; 63.05 -> 60.48 | Hurt |
| Use class-balanced instead of random sampling for PSC feature learning (ablation) | Hybrid-PSC · Long-tailed CIFAR-10 and CIFAR-100, imbalance ratios 100, 50, and 10 | top-1 accuracy: 78.82 -> 78.84; 83.86 -> 82.85; 90.06 -> 89.85; 44.91 -> 44.21; 48.93 -> 49.66; 62.37 -> 61.93 | Mixed |
Closest papers by meaning
- Decoupled Contrastive Learning for Long-Tailed Recognition (AAAI 2024 · similarity 0.92)
- Balanced Contrastive Learning for Long-Tailed Visual Recognition (CVPR 2022 · similarity 0.91)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.90)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.89)
- I-Con: A Unifying Framework for Representation Learning (ICLR 2025 · similarity 0.87)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · 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.