Research map / Label-efficient and robust learning / Recognition and long-tail
Feature Space Augmentation for Long-Tailed Data
Peng Chu, Xiao Bian, Shaopeng Liu, Haibin Ling
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 Phase-II feature space augmentation | ResNet-18 baseline · Long-tailed CIFAR-10 (IM 200) | top-1 accuracy (%): 70.22 -> 77.06 | Improved |
| add Phase-II feature space augmentation | ResNet-18 baseline · Long-tailed CIFAR-100 (IM 100) | top-1 accuracy (%): 43.65 -> 46.57 | Improved |
| add Phase-II feature space augmentation | ResNet-18 baseline · Long-tailed CIFAR-10 (IM 10) | top-1 accuracy (%): 90.73 -> 91.75 | Improved |
| add class-balanced loss (beta=0.9) to baseline | ResNet-18 · Long-tailed CIFAR-10 (IM 200) | top-1 accuracy (%): 70.22 -> 69.16 | No clear effect |
| add class-balanced loss (beta=0.999) to baseline | ResNet-18 · Long-tailed CIFAR-100 (IM 100) | top-1 accuracy (%): 43.65 -> 32.19 | Hurt |
| add focal loss (gamma=1.0) to baseline | ResNet-18 · Long-tailed CIFAR-100 (IM 200) | top-1 accuracy (%): 38.87 -> 40.47 | Improved |
| add focal loss in Phase-I to feature augmentation (Ours+FL) | Ours · ImageNet-LT (overall) | top-1 accuracy (%): 35.2 -> 35.3 | No clear effect |
| add focal loss in Phase-I to feature augmentation (Ours+FL) | Ours · Places-LT (overall) | top-1 accuracy (%): 36.4 -> 36.0 | Hurt |
| add feature augmentation on iNaturalist | ResNet-101 baseline · iNaturalist 2018 | top-1 accuracy (%): 65.19 -> 68.39 | Improved |
| add feature augmentation on iNaturalist | ResNet-50 baseline · iNaturalist 2017 | top-1 accuracy (%): 60.50 -> 61.96 | Improved |
| set head class sample ratio h_r near 0.95 instead of lower or higher (ablation) | Ours · long-tailed CIFAR | top-1 accuracy (%) | Improved |
| apply feature augmentation after Block4 instead of Block2 or Block3 (ablation) | Ours ResNet-18 · long-tailed CIFAR | top-1 accuracy (%) | Improved |
Closest papers by meaning
- Class-Balanced Loss Based on Effective Number of Samples (CVPR 2019 · similarity 0.89)
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.89)
- CUDA: Curriculum of Data Augmentation for Long-tailed Recognition (ICLR 2023 · similarity 0.89)
- Geometry of Long-Tailed Representation Learning: Rebalancing Features for Skewed Distributions (ICLR 2025 · similarity 0.89)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.87)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.86)
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.
