Research map / Label-efficient and robust learning / Recognition and long-tail
Unbiased Manifold Augmentation for Coarse Class Subdivision
Baoming Yan, Ke Gao, Bo Gao, Lin Wang, Jiang Yang, Xiaobo Li
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 UMA (same-class layer-switch synthesis, UMA A+B) over best prior augmentation | best baseline augmentation · 11 CCS tasks (32 samples), average | accuracy (%): 7.93% average improvement | Improved |
| add UMA++ (cross-class synthesis plus robust learning) over best prior augmentation | best baseline augmentation · 11 CCS tasks (32 samples), average | accuracy (%): 10.37% average improvement | Improved |
| add UMA C/D cross-class synthesis with progressive robust learning (UMA++ vs UMA) (ablation) | UMA · 11 CCS tasks (32 samples), average | accuracy (%): 79.37 -> 81.81 | Improved |
| add UMA C/D cross-class synthesis with progressive robust learning (UMA++ vs UMA) (ablation) | UMA · CelebA Eyeglasses (32 samples) | accuracy (%): 90.99 -> 93.01 | Improved |
| add UMA++ over best prior augmentation on eyeglasses recognition | best baseline augmentation · CelebA Eyeglasses (32 samples) | accuracy (%): 14.15% over best method (12.13% for UMA) | Improved |
| add UMA++ to ResNet-18 baseline under severe bias | ResNet-18 baseline · CelebA gender with eyeglasses bias | accuracy (%): 66.02 -> 82.81 | Improved |
| UMA++ vs state-of-the-art augmentation under severe bias | state-of-the-art augmentation · CelebA gender with eyeglasses bias | accuracy (%): 9.37% higher | Improved |
| add UMA++ to SPS FGVC method | SPS · CelebA gender with eyeglasses bias | accuracy (%): 72.27 -> 83.59 | Improved |
| add UMA++ to ProtoTree FGVC method | ProtoTree · CelebA gender with eyeglasses bias | accuracy (%): 66.80 -> 84.38 | Improved |
| add UMA++ to CAL FGVC method | CAL · CelebA gender with eyeglasses bias | accuracy (%): 73.05 -> 84.77 | Improved |
| use full training data instead of 32 samples, UMA vs others | UMA · sub-class recognition tasks (full training data) | accuracy (%): 2% improvement | Improved |
| use 12.5% of training data, UMA vs others | UMA · smiling and heavy makeup+smiling recognition | accuracy (%): up to 7% (smiling), 9.8% (CA#1) | Improved |
Closest papers by meaning
- Minimalistic Unsupervised Representation Learning with the Sparse Manifold Transform (ICLR 2023 · similarity 0.87)
- Curvature Enhanced Data Augmentation for Regression (ICML 2025 · similarity 0.86)
- Unsupervised Manifold Linearizing and Clustering (ICCV 2023 · similarity 0.86)
- Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds (CVPR 2025 · similarity 0.84)
- First-Order Manifold Data Augmentation for Regression Learning (ICML 2024 · similarity 0.84)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · similarity 0.84)
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.
