AI Paper Atlas

Research map / Detection and segmentation / CNN features

Interactive Segmentation with Elaborate Focus Prior

Kangpeng Hu, Yinghui Sun, Tao Wang, Weihao Zhang, Quansen Sun

ICML 2026 · Detection and segmentation › CNN features · approach: Context modeling for semantic segmentation

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 focus view predictor (FVP) (ablation)FCFI baseline (ViT-B) · SBDNoC@85 and NoC@90: 0.17 for NoC%85 and 0.28 for NoC%90; 3.49 -> 3.32; 5.83 -> 5.55Improved
add focus view predictor (FVP) (ablation)FCFI baseline (ViT-B) · BerkeleyNoC@85 and NoC@90: 0.09 for NoC%85 and 0.08 for NoC%90; 1.49 -> 1.40; 1.97 -> 1.91Improved
add triple-affinity correction (TAC) (ablation)FCFI baseline (ViT-B) · SBDNoC@85 and NoC@90: 0.1 NoC%85 and 0.12 NoC%90; 3.49 -> 3.39; 5.83 -> 5.71Improved
add clicked-with-focus integration (CFI: FCL and FA) (ablation)EFPNet · SBDNoC@85 and NoC@90: 0.12 NoC%85 and 0.12 NoC%90Improved
add all components (FVP, TAC, FCL, FA) (ablation)FCFI baseline (ViT-B) · SBDNoC@90: 5.83 -> 5.27Improved
use learnable TAC affinity weights instead of average mixture 0.5/0.5/0.5 (ablation)TAC · SBDNoC@90: 5.31 -> 5.27Improved
use learnable weights instead of pixel-only affinity (lambda_P=1) (ablation)TAC · BerkeleyNoC@90: 2.08 -> 1.87; up to 0.29 NoC%90 gainImproved
use mask-only affinity instead of pixel-only affinity (ablation)TAC · SBDNoC@90: 5.69 -> 5.44Improved
use EFPNet instead of FCFI with same backbone and dataFCFI · GrabCut/Berkeley/DAVIS/SBDNoC: around 0.2 to 0.4 NoCImproved
use EFPNet instead of FCFI (ResNet-101)FCFI · BerkeleyFIoU: 0.117 -> 0.609Improved
use EFPNet instead of FCFI (ResNet-101)FCFI · ResNet-101 efficiencyseconds per click: 0.023 -> 0.029Hurt
use EFPNet instead of FocusCut on GrabCutFocusCut · GrabCutNoC@85: 1.46 -> 1.54Hurt

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