Research map / Label-efficient and robust learning / Recognition and long-tail
Geometric Prior Guided Feature Representation Learning for Long-Tailed Classification
Yanbiao Ma, Licheng Jiao, Fang Liu, Shuyuan Yang, Xu Liu, Puhua Chen
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 third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 50.7 -> 50.9 | Improved |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · CIFAR-10-LT (IF=200) | top-1 accuracy (%): 79.6 -> 79.8 | Improved |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · ImageNet-LT (overall) | top-1 accuracy (%): 55.2 -> 55.5 | Improved |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · ImageNet-LT (Head) | top-1 accuracy (%): 65.1 -> 65.4 | Improved |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · ImageNet-LT (Tail) | top-1 accuracy (%): 38.3 -> 37.8 | Hurt |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · iNaturalist 2018 (overall) | top-1 accuracy (%): 72.4 -> 72.6 | Improved |
| add third stage (feature extractor fine-tuning) to decoupled training (ablation) | FUR · iNaturalist 2018 (Tail) | top-1 accuracy (%): 73.7 -> 73.1 | Hurt |
| add a fourth training phase to retune tail feature augmentation (ablation) | FUR · CIFAR-10-LT and CIFAR-100-LT | top-1 accuracy (%) | No clear effect |
| replace GCL feature clouds with geometry-guided feature uncertainty (FUR) | FUR vs GCL · CIFAR-10-LT and CIFAR-100-LT (IF=100) | top-1 accuracy (%): outperforms GCL by 1% and 2.2% | Improved |
| use prior-guided tail recovery instead of random foreground pasting (CMO) | FUR-Decoupled vs CMO · CIFAR-100-LT (IF=10) | top-1 accuracy (%): outperforms the combined CMO by 1.6% | Improved |
| use prior-guided tail recovery instead of random foreground pasting (CMO) | FUR-Decoupled vs CMO · ImageNet-LT and iNaturalist 2018 (Head) | top-1 accuracy (%): outperforms CMO by 3.4% and 4.8% | Improved |
Closest papers by meaning
- Geometry of Long-Tailed Representation Learning: Rebalancing Features for Skewed Distributions (ICLR 2025 · similarity 0.92)
- Curvature-Balanced Feature Manifold Learning for Long-Tailed Classification (CVPR 2023 · similarity 0.90)
- Invariant Feature Learning for Generalized Long-Tailed Classification (ECCV 2022 · similarity 0.89)
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.87)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · 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.
