Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Calibrated Hyper-Sphere Representation via Distribution Overlap Coefficient for Long-Tailed Learning
Hualiang Wang, Siming Fu, Xiaoxuan He, Hangxiang Fang, Zuozhu Liu, Haoji Hu
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 L_icd inter-class discrepancy loss (ablation) | vMF classifier · ImageNet-LT | few-shot accuracy: 26.9 -> 31.7 | Improved |
| add L_icd inter-class discrepancy loss (ablation) | vMF classifier · ImageNet-LT | overall accuracy: 52.2 -> 52.9 | Improved |
| add L_cfc class-feature consistency loss (ablation) | vMF classifier · ImageNet-LT | overall accuracy: 52.2 -> 53.1 | Improved |
| add L_cfc class-feature consistency loss (ablation) | vMF classifier · ImageNet-LT | few-shot accuracy: 26.9 -> 27.9 | Mixed |
| add both L_icd and L_cfc (ablation) | vMF classifier · ImageNet-LT | overall accuracy: 52.2 -> 53.5 | Improved |
| use cosine baseline replaced by vMF classifier with both losses (ablation) | cosine classifier · ImageNet-LT | overall accuracy: 51.8 -> 53.5 | Improved |
| raise loss weight lambda from 0.2 to 0.4 (ablation) | vMF classifier + both losses · ImageNet-LT | overall accuracy: 53.5 -> 52.6 | Hurt |
| lower loss weight lambda from 0.2 to 0.1 (ablation) | vMF classifier + both losses · ImageNet-LT | overall accuracy: 53.5 -> 53.2 | Hurt |
| add post-training calibration to linear classifier (ablation) | linear classifier · ImageNet-LT | overall accuracy: 43.2 -> 48.3 | Improved |
| add post-training calibration to tau-norm classifier (ablation) | tau-norm classifier · ImageNet-LT | overall accuracy: 48.6 -> 53.0 | Improved |
| add post-training calibration to causal classifier (ablation) | causal classifier · ImageNet-LT | overall accuracy: 49.0 -> 50.9 | Improved |
| add post-training calibration to vMF classifier (ablation) | vMF classifier · ImageNet-LT | overall accuracy: 53.5 -> 53.7 | Improved |
Closest papers by meaning
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.87)
- Feature Space Augmentation for Long-Tailed Data (ECCV 2020 · similarity 0.86)
- Improving Calibration for Long-Tailed Recognition (CVPR 2021 · similarity 0.86)
- Geometric Prior Guided Feature Representation Learning for Long-Tailed Classification (IJCV 2024 · similarity 0.86)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.85)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.85)
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.
