AI Paper Atlas

Research map / Detection and segmentation / CNN features

RepPoints: Point Set Representation for Object Detection

Ze Yang, Shaohui Liu, Han Hu, Liwei Wang, Stephen Lin

ICCV 2019 · Detection and segmentation › CNN features · approach: Box regression and loss design · 1,328 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
replace bounding box with RepPoints representation (ablation)two-stage detector (ResNet-50) · COCO minivalAP: +2.1 mAPImproved
replace bounding box with RepPoints representation (ablation)two-stage detector (ResNet-101) · COCO minivalAP: +2.0 mAPImproved
add object localization loss to RepPoints learning (ablation)RPDet · COCO minivalAP: 33.8 -> 38.3Improved
add recognition loss feedback to RepPoints learning (ablation)RPDet · COCO minivalAP: 37.6 -> 38.3Improved
add recognition loss feedback to bounding box detector (ablation)bounding box detector · COCO minivalAP: 36.2 -> 36.2No clear effect
use center point instead of single anchor as initial representation (ablation)bounding box detector · COCO minivalAP: 36.2 -> 37.3Improved
use center point instead of single anchor as initial representation (ablation)RepPoints detector · COCO minivalAP: 36.9 -> 38.3Improved
use moment-based instead of min-max pseudo box conversion (ablation)RPDet · COCO minivalAP: 38.2 -> 38.3No clear effect
add deformable RoI pooling on first-stage proposals (ablation)bounding box detector · COCO minivalAP: 36.2 -> 36.9Improved
add deformable RoI pooling on first-stage proposals (ablation)RPDet · COCO minivalAP: 38.3 -> 39.1Improved
replace YOLO-like anchor-free detector with RepPointsanchor-free detector (ResNet-50) · COCO minivalAP: 33.9 -> 38.3Improved
replace multi-anchor RetinaNet with RPDetResNet-50 detector · COCO minivalAP: 35.7 -> 38.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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