AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

RAU: Reference-based Anatomical Understanding with Vision-Language Models

Yiwei Li, Yikang Liu, Jiaqi Guo, Lin Zhao, Zheyuan Zhang, Xiao Chen, Boris Mailhe, Ankush Mukherjee, Terrence Chen, Shanhui Sun

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Multimodal LLM benchmarks

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
train with RL (GRPO 2400 steps) instead of vanillaQwen2.5VL-7B · RAOS-test-CT VQAlabeling accuracy: 16.62% to 70.68%Improved
GRPO instead of SFT (5 epochs)Qwen2.5VL-7B · RAOS-test-CT VQAlabeling accuracy: 64.11% -> 70.68%Improved
add SFT then evaluate OOD Arcade vs GRPOQwen2.5VL-7B · Arcade VQAlabeling accuracy: 33.92% -> 48.37%Improved
add global bounding-box matching (RL-Bbox) over RL-VQAVLM · RAOS labelinglabeling accuracy: 74.68% to 78.16%Improved
bounding-box matching on structurally complex dataVLM · Arcade labelinglabeling accuracy: 64.37% -> 41.09%Hurt
add SAM2 segmentation (VLM+SAM2) over RL-VQAVLM · RAOS labelinglabeling accuracy: 74.68% -> 89.38%Improved
add SAM2 segmentation (VLM+SAM2) over RL-VQAVLM · LERA labelinglabeling accuracy: 54.57% -> 61.87%Improved
VLM+SAM2 vs strongest SAM2 baseline MedSAM2-Memory-Ref-SFTRAU · four datasets averageDice: 0.31 to 0.71Improved
VLM+SAM2 vs strongest baselineRAU · four datasets averagegIoU: 0.21 to 0.50Improved
add RL stage on top of SFT VLM (ablation)RAU · Arcade segmentationDice: 0.4370 -> 0.6754Improved
add RL-VQA initialization instead of SFT-VQA (ablation)VLM+SAM2 · LERA labelinglabeling accuracy: 33.95% -> 61.87%Improved
add RL-VQA initialization instead of SFT-VQA (ablation)VLM+SAM2 · CAMUS labelinglabeling accuracy: 76.36% -> 95.41%Improved

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