AI Paper Atlas

Research map / Detection and segmentation / CNN features

Improving Fine-Grained Visual Recognition in Low Data Regimes via Self-Boosting Attention Mechanism

Yangyang Shu, Baosheng Yu, Haiming Xu, Lingqiao Liu

ECCV 2022 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 36 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 SAM to ResNet-50ResNet-50 · Stanford Cars (100 categories, 10% labels)accuracy: 36.56% -> 47.42% (up 10.86%)Improved
add SAM to ResNet-50ResNet-50 · CUB-200-2011 (200 categories, 10% labels)accuracy: 36.99% -> 40.24% (up 3.25%)Improved
add SAM to ResNet-50ResNet-50 · FGVC-Aircraft (100 categories, 10% labels)accuracy: 43.52% -> 46.73% (up 2.21%)Improved
add SAM to FBP (SAM bilinear)FBP · CUB-200-2011 (50 categories, 10% labels)accuracy: 42.82% -> 50.46% (up 7.64%)Improved
add SAM to FBP (SAM bilinear)FBP · Stanford Cars (196 categories, 10% labels)accuracy: 40.13% -> 43.19% (up 3.06%)Improved
add SAM to DBTNet-50DBTNet-50 · Stanford Cars (all categories, 10% labels)accuracy: 39.48% -> 42.47%Improved
add SAM to ResNet-50, full labelsResNet-50 · FGVC-Aircraft (100%)accuracy: 87.13% -> 86.74%Hurt
add SAM to ResNet-50, full labelsResNet-50 · CUB-200-2011 (100%)accuracy: 81.34% -> 81.62%No clear effect
add SAM to DBTNet-50, full labelsDBTNet-50 · CUB-200-2011 (100%)accuracy: 86.04% -> 87.26%Improved
add SAM to DBTNet-50, full labelsDBTNet-50 · Stanford Cars (100%)accuracy: 94.32% -> 94.18%No clear effect
use more linear projections K in SAM-Bilinear and FBP (ablation)SAM bilinear · Stanford Cars (15% labels)accuracyMixed
use few labels instead of many: gains shrink as label proportion risesSAM · three FGVC datasetsaccuracyImproved

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