AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

OraPO: Oracle-educated Reinforcement Learning for Data-efficient and Factual Radiology Report Generation

Zhuoxiao Chen, Hongyang Yu, Ying Xu, Yadan Luo, Long Duong, Yuan-Fang Li

CVPR 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 2 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 FactS reward in place of accuracy reward (ablation)GRPO · CheXpert Plusmacro F1: 0.089 -> 0.291Improved
add FactS reward in place of accuracy reward (ablation)GRPO · CheXpert Plusmacro recall: 0.162 -> 0.605Improved
add oracle DPO step (OraPO) on top of GRPO+FactS (ablation)GRPO + FactS · CheXpert Plusmacro F1: 0.291 -> 0.341Improved
add oracle DPO step (OraPO) on top of GRPO+FactS (ablation)GRPO + FactS · CheXpert Plusmacro recall: 0.605 -> 0.832Improved
train OraPO with only 400 samples instead of FactS-only 1K (ablation)OraPO + FactS · CheXpert Plusmacro F1: 0.291 -> 0.296Improved
add SFT to GRPO+FactS instead of DPO (ablation)GRPO + FactS · CheXpert Plusmacro recall: 0.605 -> 0.176Hurt
add SFT to GRPO+FactS instead of DPO (ablation)GRPO + FactS · CheXpert Plusmacro F1: 0.291 -> 0.106Hurt
use accuracy-reward GRPO on base model (ablation)Qwen2.5-VL-3B · CheXpert Plusmacro F1: 0.034 -> 0.089Improved
use OraPO+FactS with 1K samples versus MambaXray-L with 1.27MOraPO+FactS · CheXpert Plusmacro F1: 0.335 -> 0.341Improved
use OraPO+FactS versus MambaXray-LOraPO+FactS · CheXpert Plusmacro precision: 0.377 -> 0.237Hurt
use OraPO+FactS versus best baselineOraPO+FactS · MIMIC-CXRmacro F1: 0.340 -> 0.357Improved
use OraPO+FactS versus GPT-5 ThinkingOraPO+FactS · CheXpert gold-label validation setmacro F1: 0.301 -> 0.288Hurt

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