AI Paper Atlas

Research map / Detection and segmentation / CNN features

Saliency Hierarchy Modeling via Generative Kernels for Salient Object Detection

Wenhu Zhang, Liangli Zheng, Huanyu Wang, Xintian Wu, Xi Li

ECCV 2022 · Detection and segmentation › CNN features · approach: RGB salient object detection · 19 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 Saliency Hierarchy Module with static kernels (ablation)vanilla U-Net ResNet-50 · DUTS-TEmean F-measure: 82.0% -> 85.4%Improved
add Saliency Hierarchy Module with static kernels (ablation)vanilla U-Net ResNet-50 · DUT-Omean F-measure: 73.4% -> 77.2%Improved
add Hyper Kernel Generator (ablation)B + SHM (Static) · DUT-Omean F-measure: 77.2% -> 78.4%Improved
use Erode sub-saliency supervision (ablation)SHNet · DUTS-TEmean F-measure: 1.9%Improved
use Grad-CAM instead of DT prior guidance (ablation)SHNet · DUTS-TEmean F-measure: .850 -> .854Improved
add Grad-CAM prior guidance (ablation)SHNet · DUTS-TEmean F-measure: .830 -> .854Improved
generate kernels with a transformer instead of conv (Hyper Conv) (ablation)SHNet HKG · DUTS-TEmean F-measure: .859 -> .867Improved
generate hyper-kernels with transformer instead of static kernels (ablation)SHNet HKG · DUTS-TEmean F-measure: .854 -> .867Improved
share hyper-kernels across SHMs instead of multiple transformers (ablation)SHNet HKG · DUTS-TEmean F-measure: .866 -> .867No clear effect
share hyper-kernels across SHMs instead of multiple transformers (ablation)SHNet HKG · params and GFLOPsParams (M) and GFLOPs: 61.0 -> 32.4 params; 15.7 -> 15.1 GFLOPsImproved
use number of saliency levels N=3 versus over-decomposition (ablation)SHNet · DUTS-TE and ECSSDmean F-measureMixed
use larger 352 input instead of 256SHNet ResNet-50 · DUTS-TEmean F-measure: .867 -> .883Improved

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