AI Paper Atlas

Research map / Detection and segmentation / CNN features

FCOS: Fully Convolutional One-Stage Object Detection

Zhi Tian, Chunhua Shen, Hao Chen, Tong He

ICCV 2019 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 7,269 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 center-ness branch (ablation)FCOS (ResNet-50-FPN) · COCO minivalAP: 33.5 -> 37.1Improved
compute center-ness from predicted regression vector instead of separate branch (ablation)FCOS (ResNet-50-FPN) · COCO minivalAP: 33.5 -> 33.5No clear effect
add FPN multi-level prediction (ablation)FCOS · COCO trainbest possible recall (%): 95.55 -> 98.40Improved
add FPN multi-level prediction (ablation)FCOS · COCO minivalambiguous samples (%): 23.16 -> 7.14Improved
raise NMS threshold from 0.50 to 0.60 (ablation)FCOS (P5) · COCO minivalAP: 36.4 -> 36.5No clear effect
add Group Normalization to heads (ablation)FCOS (P5, NMS 0.60) · COCO minivalAP: 36.5 -> 37.1Improved
produce P6/P7 from P5 instead of C5 (ablation)FCOS · COCO minivalAP: 36.3 -> 36.4No clear effect
move center-ness branch to regression branchFCOS · COCO minivalAP: 37.1 -> 37.4Improved
add center sampling of ground-truth boxesFCOS · COCO minivalAP: 37.4 -> 38.1Improved
add GIoU lossFCOS · COCO minivalAP: 38.1 -> 38.3Improved
normalize regression targets by FPN stridesFCOS · COCO minivalAP: 38.3 -> 38.6Improved
use FCOS (no center-ness) in place of anchor RPNFaster R-CNN RPN · COCO minivalAR100: 44.7 -> 48.0Improved

Later papers that compared with this method

62 later accepted papers report FCOS as a baseline in their results tables; 59 of them report a better number on most of the results they share with it.

54 more not listed.

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