Research map / Label-efficient and robust learning / Recognition and long-tail
CUE: Concept-Aware Multi-Label Expansion to Mitigate Concept Confusion in Long-Tailed Learning
Ruichi Zhang, Chikai Shang, Jiacheng Yang, Mengke Li, Yang Zhou, Junlong Gao, Yang Lu
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 VLM instance-level cue (ablation) | LIFT (AdaptFormer + LA) · CIFAR100-LT (IR100) | few-shot accuracy: 74.3 -> 81.6 | Improved |
| add LLM class-level cue (ablation) | LIFT (AdaptFormer + LA) · CIFAR100-LT (IR100) | few-shot accuracy: 74.3 -> 80.2 | Improved |
| add both VLM and LLM cues (ablation) | LIFT (AdaptFormer + LA) · CIFAR100-LT (IR100) | overall accuracy: 80.3 -> 82.8 | Improved |
| add both cues (ablation) | LIFT · Places-LT | overall accuracy: 51.5 -> 51.7 | Improved |
| add LLM cue to VLM cue (ablation) | CUE · Places-LT | overall accuracy: 51.6 -> 51.7 | No clear effect |
| add both cues | LIFT · ImageNet-LT | few-shot accuracy: 71.5 -> 73.0 | Improved |
| add CUE | LIFT · iNaturalist2018 | overall accuracy: 79.1 -> 79.6 | Improved |
| select VLM cue classes by top-k rather than random-k (ablation) | CUE VLM cue · CIFAR100-LT (IR100) | overall accuracy: 80.1 -> 82.8 | Improved |
| select VLM cue classes by top-k rather than last-k (ablation) | CUE VLM cue · CIFAR100-LT (IR100) | overall accuracy: 79.5 -> 82.8 | Improved |
| add CUE to VPT-shallow | VPT-shallow · CIFAR100-LT (IR100) | overall accuracy: 69.7 -> 76.5 | Improved |
| add CUE to LoRA | LoRA · CIFAR100-LT (IR100) | overall accuracy: 79.1 -> 81.2 | Improved |
| add CUE to LOS trained from scratch | LOS · CIFAR100-LT (IR100, from scratch) | overall accuracy: 26.6 -> 31.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VL-LTR — VL-LTR: Learning Class-Wise Visual-Linguistic Representation for Long-Tailed Visual Recognition (ECCV 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dual-Phase Visual-Language Pretraining and Adaptation for Long-Tailed Multi-Label Recognition (AAAI 2026 · similarity 0.89)
- Long-Tailed Multi-Label Visual Recognition by Collaborative Training on Uniform and Re-Balanced Samplings (CVPR 2021 · similarity 0.88)
- Self Supervision to Distillation for Long-Tailed Visual Recognition (ICCV 2021 · similarity 0.88)
- Category-Specific Selective Feature Enhancement for Long-Tailed Multi-Label Image Classification (ICCV 2025 · similarity 0.88)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- MLC-NC: Long-Tailed Multi-Label Image Classification Through the Lens of Neural Collapse (AAAI 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.
