AI Paper Atlas

Research map / Detection and segmentation / CNN features

Training Multi-Object Detector by Estimating Bounding Box Distribution for Input Image

Jaeyoung Yoo, Hojun Lee, Inseop Chung, Geonseok Seo, Nojun Kwak

ICCV 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 5 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
use Cauchy instead of Gaussian distribution (ablation)MDOD · COCO val2017AP: 32.7 -> 33.8Improved
add level-scale operation (ablation)MDOD · COCO val2017AP: 32.9 -> 33.8Improved
add center-limit operation (on top of ltrb) (ablation)MDOD · COCO val2017AP: 32.3 -> 32.9Improved
use ltrb-transformation instead of xywh (ablation)MDOD · COCO val2017AP: 32.8 -> 33.8Improved
increase N_roi from N_gt x1 to x5 (ablation)MDOD · COCO val2017AP: 33.8 -> 33.9No clear effect
replace conventional anchor-based training with density-estimation trainingResNet50-FPN 320x320 · COCO test-devAP: 30.1 -> 33.9Improved
replace conventional anchor-based training with density-estimation trainingResNet101-FPN 512x512 · COCO test-devAP: 36.6 -> 40.0Improved
use MDOD training instead of EfficientDet trainingEfficient-D1 640x640 · COCO test-devAP: 39.6 -> 40.5Improved
use MDOD training instead of baselineResNet50-FPN 512x512 · COCO test-devFPS: 37.5 -> 43.5Improved
use MDOD on variable-size input (short-800) versus ATSSResNet-101 FPN · COCO test-dev short-800AP: 43.6 -> 42.2Hurt

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