AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

BAR — A Reinforcement Learning Agent for Bounding-Box Automated Refinement

Morgane Ayle, Jimmy Tekli, Julia El-Zini, Boulos El-Asmar, Mariette Awad

AAAI 2020 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Intrinsically motivated exploration · 12 citations (OpenAlex)

Read the paper

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
apply BAR-DRL correction to PASCAL aeroplane detector proposalsHierarchical detector · PASCAL VOC aeroplaneAP@50 (%): 7.9 -> 23.5Improved
apply BAR-DRL to human-annotator-biased boxesFA-Kit Middle · Exp. 3AP@70: 0.3 -> 93.1Improved
apply BAR-DRL to human-annotator-biased boxesFA-Kit Left · Exp. 3AP@70: 100 -> 96.1Hurt
apply BAR-DRL to RetinaNet proposalsRetinaNet · Exp. 4 StickersAP@75: 10.8 -> 64.2Improved
apply BAR-DRL to template matching proposalsTemplate matching · Exp. 5AP@50: 25.01 -> 88.5 (FA-Kit Right); 5.06 -> 8.9 (Stickers)Mixed
apply BAR-CB (online contextual bandit) instead of noneExp. 1 distribution init · Exp. 1 FA-Kit RightAP@50: 84.64 -> 81.0Hurt
BAR-CB instead of BAR-DRLBAR · Exp. 3 FA-Kit MiddleAP@70: 93.1 -> 50.8Hurt
raise beta from 0.70 to 0.80 (ablation)BAR-DRL · Exp. 1 FA-Kit Right, 50 imagesAP@70: 39.1 -> 52.5Improved
raise beta from 0.70 to 0.75/0.80 (ablation)BAR-DRL · Exp. 1 FA-Kit Middle, 50 imagesAP@50: 85.6 -> 79.5 (beta 0.75 -> 0.80)Hurt
increase training images from 50 to 150 at beta 0.80 (ablation)BAR-DRL · Exp. 1 FA-Kit RightAP@70: 52.5 -> 82.1Improved

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