AI Paper Atlas

Research map / Detection and segmentation / CNN features

SCNet: Learning Semantic Correspondence

Kai Han, Rafael S. Rezende, Bumsub Ham, Kwan-Yee K. Wong, Minsu Cho, Cordelia Schmid, Jean Ponce

ICCV 2017 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 154 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 geometric consistency term (AG) to appearance-only SCNet (ablation)SCNet-A · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)mean PCK@0.1: 66.3 -> 69.7Improved
add the AG+ geometric variant over AG (ablation)SCNet-AG · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)mean PCK@0.1: 69.7 -> 72.2Improved
fine-tune all VGG16 layers instead of only FC layers (ablation)SCNet · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)PCKNo clear effect
add geometry (AG) on transfer to PF-WILLOW (ablation)SCNet-A · PF-WILLOW (trained on PF-PASCAL)PCK@0.1: 0.725 -> 0.721No clear effect
add AG+ geometry on transfer to PF-WILLOW (ablation)SCNet-A · PF-WILLOW (trained on PF-PASCAL)PCK@0.1: 0.725 -> 0.704Hurt
replace HOG with SCNet-A learned features in LOM matching (ablation)LOM · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)matching performanceImproved
retrain UCN on PF-PASCAL instead of raw network (ablation)UCN · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)mean PCK@0.1: 36.0 -> 55.6Improved
use SCNet-AG+ instead of LOM with SCNet features for speedSCNet-AG+ · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)runtime: 3.15 s -> 0.33 sImproved
use RP proposals instead of uniform sampling (ablation)SCNet-AG+ · PF-PASCAL (mean PCK@0.1, 1000 RP proposals)PCR/mIoUImproved
use learned SCNet-AG+ instead of FCSS w/PF on PASCAL PartsSCNet-AG+ · PASCAL PartsPCK@0.05: 0.29 -> 0.18Hurt
use SCNet-AG+ instead of LOM HOGSCNet-AG+ · PASCAL PartsIoU: 0.41 -> 0.48Improved

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