Research map / Label-efficient and robust learning / Recognition and long-tail
Bag of Tricks for Long-Tailed Visual Recognition with Deep Convolutional Neural Networks
Read the paperCheck 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 |
|---|---|---|---|
| Add input mixup, CAM-based balance-sampling in DRS, and fine-tuning after mixup | ResNet-32 bag of tricks · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 30.18 -> 19.97 | Helped |
| Add input mixup, CAM-based balance-sampling in DRS, and fine-tuning after mixup | ResNet-32 bag of tricks · CIFAR-100-LT, imbalance factor 100 | Top-1 error rate: 61.73 -> 52.17 | Helped |
| Add CAM-based foreground augmentation to balance-sampling during DRS (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 29.04 -> 24.63 | Helped |
| Add CAM-based foreground augmentation to balance-sampling during DRS (ablation) | ResNet-32 · CIFAR-100-LT, imbalance factor 100 | Top-1 error rate: 59.56 -> 58.27 | Helped |
| Use CAM-based balance-sampling in DRS instead of CS CE re-weighting (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 25.31 -> 24.63 | Helped |
| Combine CS CE re-weighting with CAM-based balance-sampling (ablation) | ResNet-32 two-stage training · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 24.63 -> 24.82 | Hurt |
| Apply re-weighting directly during training rather than in deferred re-balancing (ablation) | ResNet-32 · CIFAR-100-LT | Top-1 error rate: CE 61.73; CB CE 66.40; CS CE 70.92 | Hurt |
| Apply random under-sampling directly rather than during DRS (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 30.18 -> 34.14 | Hurt |
| Apply random over-sampling directly rather than during DRS (ablation) | ResNet-32 · CIFAR-100-LT, imbalance factor 100 | Top-1 error rate: 61.73 -> 67.00 | Hurt |
| Add input mixup (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 30.18 -> 26.99 | Helped |
| Add fine-tuning after input mixup training (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 26.99 -> 26.27 | Helped |
| Add fine-tuning after manifold mixup training (ablation) | ResNet-32 · CIFAR-10-LT, imbalance factor 100 | Top-1 error rate: 26.61 -> 28.88 | Hurt |
Closest papers by meaning
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.91)
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.91)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.88)
- DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge (NeurIPS 2024 · similarity 0.88)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · 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.