AI Paper Atlas

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

Learning Deep Representation for Imbalanced Classification

Chen Huang, Yining Li, Chen Change Loy, Xiaoou Tang

CVPR 2016 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 1,044 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
Use LMLE-kNN rather than ANetCelebA face attribute classification · CelebA face attributes, averagebalanced accuracy (%): 80 -> 84Improved
Use LMLE-kNN rather than HEDBSDS500 edge detection · BSDS500ODS: 0.78 -> 0.78No clear effect
Add quintuplet sampling and triple-header loss instead of triplet loss (ablation)Imbalanced face attribute classification · CelebA face attributesaverage accuracy (%): 72.43 -> 84.26Improved
Add quintuplet sampling and triple-header loss instead of triplet loss (ablation)Imbalanced edge detection · BSDS500ODS: 0.74 -> 0.78Improved
Add resampling to Softmax (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 68.07 -> 69.43Improved
Add cost-sensitive training after resampling to Softmax (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 69.43 -> 70.16Improved
Add resampling to Triplet (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 71.29 -> 71.75Improved
Add cost-sensitive training after resampling to Triplet (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 71.75 -> 72.43Improved
Add resampling to Quintuplet (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 81.31 -> 83.39Improved
Add cost-sensitive training after resampling to Quintuplet (ablation)Face attribute classification · CelebA face attributesaverage accuracy (%): 83.39 -> 84.26Improved
Add resampling to Quintuplet (ablation)Edge detection · BSDS500ODS: 0.76 -> 0.77Improved
Add cost-sensitive training after resampling to Quintuplet (ablation)Edge detection · BSDS500ODS: 0.77 -> 0.78Improved

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