AI Paper Atlas

Research map / Detection and segmentation / CNN features

CoupleNet: Coupling Global Structure With Local Parts for Object Detection

Yousong Zhu, Chaoyang Zhao, Jinqiao Wang, Xu Zhao, Yi Wu, Hanqing Lu

ICCV 2017 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 316 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 a local part-sensitive branch coupled with a global branch (CoupleNet instead of R-FCN)R-FCN · VOC07 test (07+12)mAP (%): 79.5 -> 81.7Improved
add global RoI-pooling branch to the local branch (couple with Global FCN) (ablation)Global FCN · VOC07 test (reimplementation)mAP (%): 78.5 -> 81.7Improved
add context region (2x proposal size) to the global branch (ablation)CoupleNet · VOC07 test (07+12)mAP (%): 81.7 -> 82.1Improved
add multi-scale training (ablation)CoupleNet with context · VOC07 test (07+12)mAP (%): 82.1 -> 82.7Improved
add a learned 1x1 conv scale normalization before summing the branches (ablation)CoupleNet with element-wise sum · VOC07 test (single-scale, no context)mAP (%): 81.1 -> 81.7Improved
use L2 normalization on branch outputs instead of none (ablation)CoupleNet with element-wise sum · VOC07 test (single-scale, no context)mAP (%): 81.1 -> 80.3Hurt
use element-wise product instead of sum when coupling (with L2 normalization) (ablation)CoupleNet · VOC07 test (single-scale, no context)mAP (%): 80.3 -> 63.5Hurt
use element-wise sum instead of element-wise max when coupling (1x1 conv) (ablation)CoupleNet · VOC07 test (single-scale, no context)mAP (%): 81.3 -> 81.7Improved
couple global and local branches in one network instead of ensembling two models (ablation)R-FCN reimplementation · VOC07 test (reimplementation)mAP (%): 79.4 -> 81.7Improved
ensemble Faster R-CNN and R-FCN (ablation)R-FCN reimplementation · VOC07 test (reimplementation)mAP (%): 78.6 -> 79.6Improved
add one extra prediction head (same parameter count) to R-FCN (ablation)R-FCN · VOC07 test (reimplementation)mAP (%): 78.6 -> 78.8No clear effect
add the local branch and context at a small inference costR-FCN · VOC07 test inference timems/img: 83 -> 102Hurt

Later papers that compared with this method

3 later accepted papers report CoupleNet as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 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