AI Paper Atlas

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

Breadcrumbs: Adversarial Class-Balanced Sampling for Long-Tailed Recognition

Bo Liu, Haoxiang Li, Hao Kang, Gang Hua, Nuno Vasconcelos

ECCV 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Decoupled and calibrated long-tail training · 15 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 feature back-tracking without alignment (ablation)Decoupling (cRT) · ImageNet-LT few-shot, ResNet-10accuracy: 21.5 -> 23.8Improved
add feature back-tracking without alignment (ablation)Decoupling (cRT) · ImageNet-LT many-shot, ResNet-10accuracy: 51.8 -> 50.4Hurt
add class-specific back-tracking lengths without alignment (ablation)Decoupling (cRT) · ImageNet-LT many-shot, ResNet-10accuracy: 51.8 -> 50.8No clear effect
add class alignment (Weak Breadcrumb) (ablation)Decoupling (cRT) · ImageNet-LT overall, ResNet-10accuracy: 41.4 -> 43.2Improved
add class alignment (Weak Breadcrumb) (ablation)Decoupling (cRT) · ImageNet-LT few-shot, ResNet-10accuracy: 21.5 -> 25.1Improved
use snapshots from throughout stage 1 (Strong vs Weak Breadcrumb) (ablation)Breadcrumb · ImageNet-LT overall, ResNet-10accuracy: 43.2 -> 44.0Improved
use snapshots from throughout stage 1 (Strong vs Weak Breadcrumb) (ablation)Breadcrumb · ImageNet-LT few-shot, ResNet-10accuracy: 25.1 -> 26.4Improved
add variance alignment on top of mean alignment (ablation)Breadcrumb · ImageNet-LT overall, ResNet-10accuracy: 44.0 -> 43.9No clear effect
use class-specific mean alignment instead of class-agnostic alignment (ablation)Breadcrumb · ImageNet-LT overall, ResNet-10accuracy: 38.5 -> 44.0Improved
replace Decoupling (cRT) with BreadcrumbcRT · ImageNet-LT few-shot, ResNet-10accuracy: 21.5 -> 26.4Improved
replace Decoupling (cRT) with BreadcrumbcRT · Places-LT overall, ResNet-152accuracy: 37.9 -> 39.3Improved
replace Decoupling (cRT) with BreadcrumbcRT · Places-LT many-shot, ResNet-152accuracy: 37.8 -> 40.6Improved

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