AI Paper Atlas

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

Targeted Supervised Contrastive Learning for Long-Tailed Recognition

Tianhong Li, Peng Cao, Yuan Yuan, Lijie Fan, Yuzhe Yang, Rogerio S. Feris, Piotr Indyk, Dina Katabi

CVPR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Decoupled and calibrated long-tail training · 213 citations (OpenAlex)

Check 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 changedWhereMeasuredEffect
Add uniformly distributed targets and online matching to supervised contrastive learningKCL · CIFAR-10-LT, rho=100top-1 accuracy: 77.6 -> 79.7Improved
Add uniformly distributed targets and online matching to supervised contrastive learningKCL · CIFAR-10-LT, rho=50top-1 accuracy: 81.7 -> 82.9Improved
Add uniformly distributed targets and online matching to supervised contrastive learningKCL · CIFAR-100-LT, rho=100top-1 accuracy: 42.8 -> 43.8Improved
Add uniformly distributed targets and online matching to supervised contrastive learningKCL · ImageNet-LT, alltop-1 accuracy: 51.5 -> 52.4Improved
Add uniformly distributed targets and online matching to supervised contrastive learningKCL · iNaturalist 2018, alltop-1 accuracy: 68.6 -> 69.7Improved
Use class-balanced sampling with KCL instead of instance-balanced sampling (ablation)KCL · CIFAR-10-LT, rho=100top-1 accuracy: 77.6 -> 75.5Hurt
Use class-balanced sampling with KCL instead of instance-balanced sampling (ablation)KCL · CIFAR-100-LT, rho=100top-1 accuracy: 42.8 -> 41.5Hurt
Replace online matching with fixed random class-target assignment (ablation)TSC · ImageNet-LTtop-1 accuracy: 52.4 -> 50.8Hurt
Replace online matching with fixed random class-target assignment (ablation)TSC · ImageNet-LTreasonability (R): 7.14 -> 7.81Hurt
Use KCL's balanced positive sampling rather than FCL's positive sampling (ablation)TSC · ImageNet-LT, alltop-1 accuracy: 51.8 -> 52.4Improved
Replace RIDE stage-1 training with TSCRIDE (4 experts) · ImageNet-LT, alltop-1 accuracy: 55.4 -> 56.9Improved

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