AI Paper Atlas

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

Influence-Balanced Loss for Imbalanced Visual Classification

Seulki Park, Jongin Lim, Younghan Jeon, Jin Young Choi

ICCV 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 213 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 influence-balanced fine-tuningCE baseline ResNet-32 · CIFAR-10 long-tailed rho=100accuracy: 70.87 -> 78.26Improved
add influence-balanced fine-tuningCE baseline ResNet-32 · CIFAR-100 long-tailed rho=100accuracy: 38.05 -> 42.14Improved
add influence-balanced fine-tuningCE baseline ResNet-50 · iNaturalist 2018top-1 accuracy: 57.30 -> 65.39Improved
add influence-balanced fine-tuningCE baseline ResNet-18 · Tiny ImageNet long-tailed rho=100accuracy: 38.52 -> 42.65Improved
use L1 norm instead of L2 for influence magnitude (ablation)IB loss · CIFAR-10 long-tailed rho=100accuracy: 75.67 -> 78.41Improved
use L1 norm instead of L-infinity for influence magnitude (ablation)IB loss · CIFAR-10 long-tailed rho=100accuracy: 77.23 -> 78.41Improved
add the IB weighting factor to the epsilon constant (ablation)IB loss · CIFAR-10 long-tailed rho=100accuracy: 64.91 -> 78.17Improved
use epsilon 1e-3 instead of 1e-8 (ablation)IB loss · CIFAR-10 long-tailed rho=100accuracy: 76.03 -> 78.17Improved
use epsilon 1e-3 instead of 1e-2 (ablation)IB loss · CIFAR-10 long-tailed rho=100accuracy: 77.55 -> 78.17No clear effect
combine IB with focal loss in fine-tuningIB loss · CIFAR-10 long-tailed rho=100accuracy: 78.26 -> 79.76Improved
combine IB with CB effective numberIB loss · CIFAR-10 long-tailed rho=100accuracy: 78.26 -> 78.04No clear effect
combine IB with CB effective numberIB loss · CIFAR-100 long-tailed rho=100accuracy: 42.14 -> 41.31Hurt

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