AI Paper Atlas

Research map / Video / Video-language understanding

MedGRPO: Multi-Task Reinforcement Learning for Heterogeneous Medical Video Understanding

Yuhao Su, Anwesa Choudhuri, Zhongpai Gao, Benjamin Planche, Van Nguyen Nguyen, Meng Zheng, Yuhan Shen, Arun Innanje, Terrence Chen, Ehsan Elhamifar, Ziyan Wu

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding · 7 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 cross-dataset reward normalization to GRPO (ablation)MedGRPO · MedVidBenchCVS accuracy: 0.020 -> 0.896Improved
add cross-dataset reward normalization to GRPO (ablation)MedGRPO · MedVidBenchTAG mIoU@0.3: 0.004 -> 0.216Improved
add cross-dataset reward normalization to GRPO (ablation)MedGRPO · MedVidBenchVS LLM judge: 3.805 -> 4.184No clear effect
add caption rewards (VS+RC) alongside grounding rewards (ablation)MedGRPO · MedVidBenchTAG mIoU@0.3: 0.202 -> 0.216Improved
add caption rewards alongside grounding rewards (ablation)MedGRPO · MedVidBenchSTG mIoU: 0.193 -> 0.202Improved
add caption rewards alongside grounding rewards (ablation)MedGRPO · MedVidBenchDVC F1: 0.225 -> 0.214Hurt
add medical LLM judge to caption reward (ablation)MedGRPO (VS+RC) · MedVidBenchRC LLM judge: 2.984 -> 3.235Improved
add medical LLM judge to caption reward (ablation)MedGRPO (VS+RC) · MedVidBenchVS LLM judge: 3.733 -> 3.824Improved
apply MedGRPO RL on top of SFTQwen2.5VL-7B SFT · MedVidBenchVS LLM judge: 3.596 -> 4.184Improved
apply MedGRPO RL on top of SFTQwen2.5VL-7B SFT · MedVidBenchNAP accuracy: 0.442 -> 0.405Hurt
apply MedGRPO RL on top of SFTQwen2.5VL-7B SFT · MedVidBenchCVS accuracy: 0.894 -> 0.896No clear effect
SFT on MedVidBench vs off-the-shelf Qwen2.5VL-7BQwen2.5-VL-7B · MedVidBenchCVS accuracy: 0.105 -> 0.894Improved

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