Research map / Vision-language and multimodal / Human and embodied modeling
UniMedVL: Unifying Medical Multimodal Understanding and Generation through Observation-Knowledge-Analysis
Junzhi Ning, Wei Li, Cheng Tang, Jiashi Lin, Chenglong Ma, Chaoyang Zhang, Jiyao Liu, Ying Chen, Shujian Gao, Yuandong Pu, Huihui Xu, Chenhui Gou and 15 more
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add understanding training to generation-only model (ablation) | UniMedVL · medical text-to-image (8 modalities, avg) | gFID: 108.40 -> 96.29 | Improved |
| add generation training to understanding-only model (ablation) | UniMedVL (Stage 1) · GMAI-MMBench | accuracy: 0.505 -> 0.593 | Improved |
| add generation training to understanding-only model (ablation) | UniMedVL (Stage 1) · PathVQA | accuracy: 0.367 -> 0.365 | No clear effect |
| add generation training to understanding-only model (ablation) | UniMedVL (Stage 1) · SLAKE | accuracy: 0.548 -> 0.684 | Improved |
| use three-stage progressive curriculum instead of one-stage joint training (ablation) | UniMedVL · GMAI-MMBench | accuracy: 0.535 -> 0.608 | Improved |
| use three-stage progressive curriculum instead of one-stage joint training (ablation) | UniMedVL · medical text-to-image gFID | gFID: 123.48 -> 96.29 | Improved |
| add interleaved tasks in Stage 3 (ablation) | UniMedVL · medical text-to-image gFID | gFID: 120.04 -> 96.29 | Improved |
| add distilled chain-of-thought data to understanding instructions (ablation) | UniMedVL (U-only) · PathVQA | accuracy: 0.367 -> 0.453 | Improved |
| add caption augmented generation (CAG) (ablation) | UniMedVL (G-only) · medical text-to-image gFID | gFID: 118.60 -> 108.40 | Improved |
| add high-quality instruction data in Stage 2 (ablation) | UniMedVL · SLAKE | accuracy: 0.684 -> 0.742 | Improved |
| add interleaved data in Stage 3 (ablation) | UniMedVL · OmniMedVQA | accuracy: 0.863 -> 0.858 | No clear effect |
| fine-tune FLUX VAE end-to-end on medical images (ablation) | UniMedVL · rFID across modalities | rFID | No clear effect |
Later papers that compared with this method
1 later accepted paper reports UniMedVL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- SynerMedGen: Synergizing Medical Multimodal Understanding with Generation via Task Alignment (ICML 2026 · better on 6 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning (CVPR 2026 · similarity 0.88)
- HealthGPT: A Medical Large Vision-Language Model for Unifying Comprehension and Generation via Heterogeneous Knowledge Adaptation (ICML 2025 · similarity 0.87)
- How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images (ICLR 2026 · similarity 0.87)
- MedSIGHT: Towards Grounded Visual Comprehension in Medical Large Vision-Language Models (ICML 2026 · similarity 0.87)
- RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis (NeurIPS 2025 · similarity 0.87)
- GMAI-VL & GMAI-VL-5.5M: A Large Vision-Language Model and a Comprehensive Multimodal Dataset Towards General Medical AI (AAAI 2026 · similarity 0.87)
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.
