AI Paper Atlas

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

Distribution-Balanced Loss for Multi-Label Classification in Long-Tailed Datasets

Tong Wu, Qingqiu Huang, Ziwei Liu, Yu Wang, Dahua Lin

ECCV 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning · 293 citations (OpenAlex)

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 changedWhereMeasuredEffect
add re-balanced weighting to class-aware re-sampling BCE (ablation)BCE with re-sampling · VOC-MLTtotal mAP: 75.38 -> 76.34Improved
add re-balanced weighting to class-aware re-sampling BCE (ablation)BCE with re-sampling · COCO-MLTtotal mAP: 46.97 -> 49.43Improved
add negative-tolerant regularization to R-BCE (ablation)R-BCE · VOC-MLTtotal mAP: 76.34 -> 78.65Improved
add negative-tolerant regularization to R-BCE (ablation)R-BCE · COCO-MLTtotal mAP: 49.43 -> 52.53Improved
add focal loss to DB loss (ablation)DB · VOC-MLTtotal mAP: 78.65 -> 78.94Improved
add focal loss to DB loss (ablation)DB · COCO-MLTtotal mAP: 52.53 -> 53.55Improved
add focal loss to re-sampling (ablation)RS · COCO-MLTtotal mAP: 46.97 -> 51.14Improved
add re-balanced weighting to RS-Focal (ablation)RS-Focal · VOC-MLTmedium-class mAP: drop 0.3% for medium classesHurt
add class-aware re-sampling to ERMERM · COCO-MLThead-class mAP: drops by about 0.9%Hurt
add class-level re-weighting (RW) on top of re-sampling with focal lossRS-Focal · VOC-MLTtotal mAP: 76.45 -> 71.96Hurt
add class-balanced re-weighting on top of re-sampling with focal lossRS-Focal · COCO-MLTtotal mAP: 51.14 -> 50.07Hurt
use re-balanced weighting instead of traditional re-weighting with re-sampling and focal lossRS-Focal · COCO-MLTtotal mAP: 50.07 -> 52.75Improved

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