AI Paper Atlas

Research map / Image generation / Image restoration

Learning Deep Local Features With Multiple Dynamic Attentions for Large-Scale Image Retrieval

Hui Wu, Min Wang, Wengang Zhou, Houqiang Li

ICCV 2021 · Image generation › Image restoration · approach: Dehazing and deraining · 30 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 dynamic attention (DA) head instead of 1x1 conv attention (ablation)MDA, N=8 · ROxf HardmAP: 60.6 -> 62.2Improved
add dynamic attention (DA) head instead of l2 norm attention (ablation)MDA, N=8 · ROxf HardmAP: 60.5 -> 62.2Improved
use DA instead of 1x1 conv + l2 norm attention (ablation)MDA, N=8 · RPar HardmAP: 64.8 -> 66.2Improved
add stop gradient from attention branch to backbone (ablation)MDA with DA, N=8 · ROxf HardmAP: 60.3 -> 62.2Improved
add stop gradient from attention branch to backbone (ablation)MDA with DA, N=8 · RPar HardmAP: 64.0 -> 66.2Improved
add channel mapping layer to diversity regularization (ablation)MDA, N=8 · ROxf HardmAP: 59.4 -> 62.2Improved
add channel mapping layer to diversity regularization (ablation)MDA, N=8 · RPar HardmAP: 61.2 -> 66.2Improved
add diversity regularization to channel mapping (ablation)MDA, N=8 · ROxf HardmAP: 60.2 -> 62.2Improved
add diversity regularization to channel mapping (ablation)MDA, N=8 · RPar HardmAP: 64.7 -> 66.2Improved
increase head number from 1 to 8 (ablation)MDA · ROxf-RPar Hard and MediummAP: 57.8 -> mAP rises until N=8Improved
increase head number to 10 or more (ablation)MDA · ROxf-RParmAPHurt
single head DA instead of HOW attentionMDA N=1 vs HOW · Hard, ROxfmAP: +0.9 mAP (57.8 vs HOW)Improved

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