AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Sharpen Focus: Learning With Attention Separability and Consistency

Lezi Wang, Ziyan Wu, Srikrishna Karanam, Kuan-Chuan Peng, Rajat Vikram Singh, Bo Liu, Dimitris N. Metaxas

ICCV 2019 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 28 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 inner-layer attention separation loss L_AS^in (ablation)ResNet-50 · CUB-200-2011top-1 accuracy: 81.70 -> 85.15Improved
add attention consistency loss L_AC (ablation)ResNet-50 + L_AS^in · CUB-200-2011top-1 accuracy: 85.15 -> 85.77Improved
add last-layer attention separation loss L_AS^la (ablation)ResNet-50 + L_AS^in + L_AC · CUB-200-2011top-1 accuracy: 85.77 -> 86.20Improved
use channel-weighted attention A_ch instead of Grad-CAM inside ICASC (ablation)ICASC · CUB-200-2011top-1 accuracy: 85.45 -> 86.20Improved
use channel-weighted attention A_ch instead of Grad-CAM inside ICASC (ablation)ICASC · Caltech-256 (N=60 per class)top-1 accuracy: 80.28 -> 81.32Improved
use channel-weighted attention A_ch instead of Grad-CAM inside ICASC (ablation)ICASC · ILSVRC2012 val (single crop)top-1 accuracy: 69.84 -> 69.90No clear effect
use channel-weighted attention A_ch instead of Grad-CAM inside ICASC (ablation)ICASC · PASCAL VOC 2012 classificationAP (%): 82.12 -> 83.17Improved
use A_ch instead of Grad-CAM++ as weak localization cue (ablation)DeepLab on VGG-19 attention · PASCAL VOC 2012 seg val (weak supervision)segmentation score: 51.70 -> 57.97Improved
use A_ch instead of Grad-CAM as weak localization cue (ablation)DeepLab on VGG-19 attention · PASCAL VOC 2012 seg val (weak supervision)segmentation score: 56.65 -> 57.97Improved
add ICASC training with Grad-CAM attentionResNet-110 · CIFAR-100top-1 accuracy: 72.78 -> 74.02Improved
add ICASC training with A_ch attentionResNet-110 · CIFAR-100top-1 accuracy: 72.78 -> 76.11Improved
add ICASC training with A_ch attentionResNet-18 · ILSVRC2012 val (single crop)top-1 accuracy: 69.51 -> 69.90No clear effect

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