Research map / Label-efficient and robust learning / Recognition and long-tail
BBN: Bilateral-Branch Network With Cumulative 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 |
|---|---|---|---|
| Use BBN's bilateral branches and cumulative learning | ResNet-32 · CIFAR-10-IR100 | top-1 error rate: 20.18 vs 22.97 | Helped |
| Use BBN with the 2× training schedule | ResNet-50 · iNaturalist 2018 | top-1 error rate: 30.38 vs 33.88 | Helped |
| Use a reversed sampler in the re-balancing branch (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 21.31 -> 17.82; 21.06 -> 17.82 | Helped |
| Replace reversed sampling with uniform sampling in the re-balancing branch (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 17.82 -> 21.31 | Hurt |
| Replace reversed sampling with balanced sampling in the re-balancing branch (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 17.82 -> 21.06 | Hurt |
| Use parabolic decay for α instead of equal weighting (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 21.56 -> 17.82 | Helped |
| Use parabolic decay for α instead of β-distribution sampling (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 21.75 -> 17.82 | Helped |
| Use parabolic decay rather than parabolic increment for α (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 22.70 -> 17.82 | Helped |
| Use parabolic decay rather than linear decay for α (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 18.55 -> 17.82 | Helped |
| Use parabolic decay rather than cosine decay for α (ablation) | BBN, ResNet-32 · CIFAR-10-IR50 | error rate: 18.04 -> 17.82 | Helped |
| Train representations with re-weighting rather than conventional cross-entropy | feature extractor · CIFAR-100-IR50 | error rate: 58.62 -> 63.17 | Hurt |
| Train representations with re-sampling rather than conventional cross-entropy | feature extractor · CIFAR-100-IR50 | error rate: 58.62 -> 63.71 | Hurt |
Closest papers by meaning
- Bag of Tricks for Long-Tailed Visual Recognition with Deep Convolutional Neural Networks (AAAI 2021 · similarity 0.90)
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.90)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.89)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.89)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
- MLC-NC: Long-Tailed Multi-Label Image Classification Through the Lens of Neural Collapse (AAAI 2025 · 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.