AI Paper Atlas

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

Orthogonal Projection Loss

Kanchana Ranasinghe, Muzammal Naseer, Munawar Hayat, Salman Khan, Fahad Shahbaz Khan

ICCV 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 100 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 OPL to CEResNet-56 · CIFAR-100top-1 accuracy: 72.40% -> 73.52%Improved
add OPL to CEResNet-110 · CIFAR-100top-1 accuracy: 73.79% -> 74.85%Improved
add OPL to CEResNet-50 · ImageNettop-1 accuracy: 76.15% -> 76.98%Improved
add OPL to CEResNet-18 · ImageNettop-1 accuracy: 69.91% -> 70.27%Improved
add OPL to Truncated LossResNet18 · CIFAR100 class-dependent noisy labelsaccuracy: 47.66% -> 53.94%Improved
add OPL to Madry adversarial trainingMadry et al. · CIFAR10 AutoAttackrobust accuracy: 44.04 -> 49.15Improved
add OPL to MARTMART · CIFAR10clean accuracy: 84.49 -> 84.41No clear effect
add OPL to MARTMART · CIFAR100clean accuracy: 58.90 -> 58.01Hurt
add OPL to RSCRSC · PACSaverage accuracy: 87.47 -> 88.48Improved
add OPL to RFSRFS · miniImageNet 1-shotaccuracy: 62.02 -> 63.10Improved
add OPL to SURSUR · Meta-Dataset averageaccuracy: 71.4 -> 71.9Improved
add OPL to RFS under Gaussian input noise sigma=0.1 (ablation)RFS · miniImageNet 5-shot (noisy)accuracy: 74.46 -> 73.20Hurt

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