AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

MedLoc-R1: Performance-Aware Curriculum Reward Scheduling for GRPO-Based Medical Visual Grounding

Guangjing Yang, Ziyuan Qin, Chaoran Zhang, Chenlin Du, Jinglin Wang, Wanran Sun, Zhenyu Zhang, Bing Ji, Qicheng Lao

CVPR 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 1 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 fixed IoU threshold 0.5 reward with performance-aware curriculum thresholdGRPO Qwen2.5-VL-3B · HEELA@0.8: 4.17 -> 47.35Improved
replace fixed IoU threshold 0.5 reward with performance-aware curriculum thresholdGRPO Qwen2.5-VL-3B · HAM10000A@0.5: 64.65 -> 94.46Improved
use performance-aware schedule instead of progress-only V-Triune scheduleGRPO Qwen2.5-VL-3B · TN3KA@0.5: 43.50 -> 66.18Improved
use curriculum binary thresholds instead of continuous IoU rewardGRPO Qwen2.5-VL-3B · HEELA@0.8: 11.49 -> 47.35Improved
compare MedLoc-R1 to SFTQwen2.5-VL-3B · TN3KA@0.8: 28.71 -> 29.60Improved
scale model from 3B to 7B/32BMedLoc-R1 · TN3KA@0.8: 29.60 -> 27.02 (7B), 27.52 (32B)Mixed
scale model from 3B to 7BMedLoc-R1 · HEELA@0.8: 47.35 -> 57.83Improved
add reward-check condition to threshold update criterion (ablation)MedLoc-R1 · HAM10000A@0.5: 82.33 -> 94.96Improved
add IoU margin check to update criterion (ablation)MedLoc-R1 · HAM10000A@0.5: 90.86 -> 94.96Improved
add stability (reward std) check to update criterion (ablation)MedLoc-R1 · HAM10000A@0.5: 89.72 -> 94.96Improved
use dynamic step size/P/S instead of fixed aggressive setting (ablation)MedLoc-R1 · HAM10000A@0.5: 71.54 -> 94.96Improved
use dynamic step size instead of identical constant step size (ablation)MedLoc-R1 · HAM10000A@0.5: 79.64 -> 94.96Improved

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