Research map / Label-efficient and robust learning / Recognition and long-tail
CUDA: Curriculum of Data Augmentation for Long-tailed Recognition
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 class-wise adaptive augmentation with CUDA | BCL · CIFAR-100-LT IR 100 | accuracy: 51.0 -> 52.3 ± 0.2 | Helped |
| Add class-wise adaptive augmentation with CUDA | BCL · CIFAR-100-LT IR 50 | accuracy: 54.9 -> 56.2 ± 0.4 | Helped |
| Add class-wise adaptive augmentation with CUDA | BCL · CIFAR-100-LT IR 10 | accuracy: 64.4 -> 64.6 ± 0.1 | Helped |
| Add class-wise adaptive augmentation with CUDA | BCL · ImageNet-LT | accuracy: 55.6 ± 0.2 -> 56.3 ± 0.1 | Helped |
| Add class-wise adaptive augmentation with CUDA | RIDE (3 experts) · iNaturalist 2018 | accuracy: 71.6 ± 0.1 -> 72.4 ± 0.2 | Helped |
| Use static final augmentation strengths instead of the evolving curriculum (ablation) | CUDA · CIFAR-100-LT IR 100 | accuracy | Hurt |
| Replace CUDA with hyperparameter optimization for choosing class-wise augmentation strengths (ablation) | CUDA · CIFAR-100-LT IR 100 | accuracy | Hurt |
| Increase the augmentation preset from 10 to 14 operations (ablation) | CUDA · CIFAR-100-LT IR 100 | accuracy: 41.6 -> 41.8 | Helped |
| Increase the augmentation preset from 14 to 22 operations (ablation) | CUDA · CIFAR-100-LT IR 100 | accuracy: 41.8 -> 42.7 | Helped |
| Use fixed rather than random augmentation operation order (ablation) | CUDA · CIFAR-100-LT IR 100 | accuracy: 42.7 -> 42.4 | No clear effect |
| Set the LoL threshold gamma to 0.3 instead of 0.4 (ablation) | BS + CUDA · ImageNet-LT | accuracy: 51.59 -> 51.42 | Hurt |
| Set the LoL threshold gamma to 0.5 instead of 0.4 (ablation) | BS + CUDA · ImageNet-LT | accuracy: 51.59 -> 51.38 | Hurt |
Closest papers by meaning
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.88)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
- Exploring Weight Balancing on Long-Tailed Recognition Problem (ICLR 2024 · similarity 0.87)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.87)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 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.