Research map / Label-efficient and robust learning / Recognition and long-tail
Confusion-Aware Spectral Regularizer for Long-Tailed Recognition
Ziquan Zhu, Gaojie Jin, Hanruo Zhu, Si-Yuan Lu, Yunxiao Zhang, Zeyu Fu, Ronghui Mu, Guoqiang Zhang, Zhao Sun, Yuhang Xia, Jiaxing Shang, Xiang Li and 2 more
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 |
|---|---|---|---|
| remove frequency-aware class weight Lambda (ablation) | CAR · ImageNet-LT (from scratch) | overall accuracy: 54.39 -> 57.48 (with Lambda) | Improved |
| remove frequency-aware class weight Lambda (ablation) | CAR · CIFAR100-LT (from scratch) | overall accuracy: 49.62 -> 51.85 (with Lambda) | Improved |
| add EMA confusion estimator (ablation) | CAR · ImageNet-LT (from scratch) | overall accuracy: 55.77 -> 57.48 | Improved |
| add EMA confusion estimator (ablation) | CAR · CIFAR100-LT (from scratch) | overall accuracy: 50.20 -> 51.85 | Improved |
| add CAR regularizer to cross-entropy training | ViT-Small · ImageNet-LT (from scratch) | overall accuracy: 46.51 -> 57.48 | Improved |
| combine CAR with ConCutMix | CAR · ImageNet-LT (from scratch) | overall accuracy: 57.48 -> 60.07 | Improved |
| add CAR to long-tailed augmentation methods (ReMix, MetaSAug, CMO, SAFA, ConCutMix) | CAR · ImageNet-LT / CIFAR100-LT, IF 50/100/200 | overall accuracy: e.g. 57.48 -> 60.07 (ImageNet-LT IF=100 + ConCutMix) | Improved |
| CAR vs LOS on worst-class test accuracy | LOS · ImageNet-LT | worst-class test accuracy: 10 -> 18 | Improved |
| CAR vs LOS under pre-trained ViT-Small | LOS · iNaturalist (pretrained) | overall accuracy: 83.02 -> 83.74 | Improved |
| CAR vs best baseline head accuracy | LOS · ImageNet-LT (from scratch) | head accuracy: 70.79 -> 67.00 | Hurt |
| CAR across backbones ViT-Tiny/Base/Large, ResNet, Swin | best baseline · ImageNet-LT | top-1 accuracy: ViT-Large 68.63 -> 69.26 | Improved |
Closest papers by meaning
- Deep Long-Tailed Learning: A Survey (TPAMI 2023 · similarity 0.90)
- Decoupling Representation and Classifier for Long-Tailed Recognition (ICLR 2020 · similarity 0.90)
- U-MOF: Uncertainty-Guided Parameter-Efficient Multi-Objective Fine-Tuning for Long-Tailed Recognition (NeurIPS 2026 · similarity 0.89)
- Decoupled Optimisation for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.88)
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.
