AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Long-Tailed Classification with Multi-Granularity Semantics

Yuting Liu, Liu Yang, Yu Wang

ICCV 2025 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive learning for imbalanced data · 2 citations (OpenAlex)

Read the paper

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 changedWhereMeasuredEffect
add LLM-label-semantic multi-granularity contrastive learning versus ConCutMixSKCL (ResNet-32) · CIFAR-100-LT (imbalance 100)Top-1 accuracy (%): 53.16 -> 54.02Improved
add SKCL on top of the BCL baselineResNet-50 · ImageNet-LTFew-shot accuracy (%): 36.5 -> 41.5Improved
add SKCL improves head classes as well as tailSKCL (ResNet-32) · CIFAR-100-LT (imbalance 100)Many-shot accuracy (%): 67.4 -> 68.3Improved
add coarse-to-fine (vertical) transfer only (ablation)SKCL · CIFAR-100-LT (imbalance 100)overall accuracy (%): 51.9 -> 53.1Improved
add same-level (horizontal) transfer only (ablation)SKCL · CIFAR-100-LT (imbalance 100)overall accuracy (%): 51.9 -> 53.2Improved
combine vertical and horizontal transfer (ablation)SKCL · CIFAR-100-LT (imbalance 100)overall accuracy (%): 53.2 -> 54.0Improved
use LLM-generated label descriptions instead of original label semantics for similaritySKCL · CIFAR-10-LTsimilarity matrix (qualitative)Improved
restrict transfer to Top-K similar classesSKCL · long-tailed recognitiontraining behavior (qualitative)Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy