AI Paper Atlas

Research map / Detection and segmentation / CNN features

Towards Bridging Semantic Gap to Improve Semantic Segmentation

Yanwei Pang, Yazhao Li, Jianbing Shen, Ling Shao

ICCV 2019 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation · 139 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 parallel feature pyramid (PFP) without SeEM (ablation)ResNet50 baseline · PASCAL VOC 2012 valmIoU: 70.9 -> 72.7Improved
add PFP without SeEM (ablation)ResNet101 baseline · PASCAL VOC 2012 valmIoU: 74.9 -> 75.7Improved
embed three SeEM modules in PFP (ablation)PFP with ResNet101 · PASCAL VOC 2012 valmIoU: 75.7 -> 76.9Improved
add a single SeEM-FS module to the baseline (ablation)ResNet101 baseline · PASCAL VOC 2012 valmIoU: 74.9 -> 75.9Improved
embed three SeEM-FS modules instead of SeEM in PFP (ablation)PFP with ResNet101 · PASCAL VOC 2012 valmIoU: 76.9 -> 77.5Improved
add boundary refinement by skip connection (ablation)PFPc · PASCAL VOC 2012 valmIoU: 77.4 -> 77.7Improved
use boundary attention module (BAM) instead of skip connection (ablation)PFPc · PASCAL VOC 2012 valmIoU: 77.4 -> 78.1Improved
enhance Block2 features with SeEM before prediction (ablation)SeENet features · PASCAL VOC 2012 valmIoU: 50.1 -> 62.4Improved
combine Block3 with SeEM-enhanced Block2 instead of raw Block2 (ablation)SeENet features · PASCAL VOC 2012 valmIoU: 71.5 -> 73.5Improved
use BAM instead of no boundary refinement, 1 pixel trimap (ablation)SeENet · PASCAL VOC 2012 valboundary mIoU: 50.5 -> 56.1Improved
increase training crop size from 320 to 512SeENet · PASCAL VOC 2012 testmIoU: +1.3%Improved
add COCO pre-training and multi-scale testingSeENet · PASCAL VOC 2012 testmIoU: 83.8 -> 86.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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