@inproceedings{jiang-etal-2026-omniv,
title = "{O}mni{V}-{M}ed: A Scalable Training Recipe for Unified Medical Vision-Language Understanding",
author = "Jiang, Songtao and
Wang, Yuan and
Zhou, Chenyi and
Hu, Tianxiang and
Niu, Fucheng and
Liu, Zuozhu",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.275/",
pages = "6263--6279",
ISBN = "979-8-89176-459-0",
abstract = "Practical deployment of medical vision-language models requires a single model that interprets heterogeneous clinical visual inputs: 2D images, 3D volumes, and medical videos. Yet how to train such unified models remains underexplored, as these modalities differ in spatial-temporal structure, token length, and data availability. We present OmniV-Med, a scalable training recipe built on four empirical findings. First, broad modality coverage requires balancing modality diversity with task-level supervision, motivating OmniV-Med-Instruct, a 252K corpus spanning 14 medical modalities and 11 clinical tasks. Second, we identify a modality-ordering effect: introducing 3D and video too early destabilizes optimization and weakens 2D grounding, motivating a progressive curriculum that first builds 2D alignment before extending to spatial-temporal inputs. Third, we observe an emergent spatiotemporal transfer effect: once strong 2D patch representations are in place, lightweight rotary position adaptation extrapolates them to 3D slices and video frames, letting a single 2D ViT replace modality-specific encoder stacks. Fourth, the imbalance in sequence length across modalities and the redundancy of long volumes and videos hinder scaling; our Redundancy-Guided Token Pruning (RGTP) exploits inter-slice and inter-frame redundancy to cut visual tokens by 60{\%} while preserving performance. Across nine 2D, 3D, and video medical benchmarks, OmniV-Med achieves robust performance. Extensive component-wise ablations systematically refine the design space, yielding a simple yet efficient and scalable recipe for training high-performance unified Med-VLMs. Code is released at \url{https://github.com/ZJUI-AI4H/Hulu-Med}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jiang-etal-2026-omniv">
<titleInfo>
<title>OmniV-Med: A Scalable Training Recipe for Unified Medical Vision-Language Understanding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Songtao</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenyi</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianxiang</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fucheng</namePart>
<namePart type="family">Niu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zuozhu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Practical deployment of medical vision-language models requires a single model that interprets heterogeneous clinical visual inputs: 2D images, 3D volumes, and medical videos. Yet how to train such unified models remains underexplored, as these modalities differ in spatial-temporal structure, token length, and data availability. We present OmniV-Med, a scalable training recipe built on four empirical findings. First, broad modality coverage requires balancing modality diversity with task-level supervision, motivating OmniV-Med-Instruct, a 252K corpus spanning 14 medical modalities and 11 clinical tasks. Second, we identify a modality-ordering effect: introducing 3D and video too early destabilizes optimization and weakens 2D grounding, motivating a progressive curriculum that first builds 2D alignment before extending to spatial-temporal inputs. Third, we observe an emergent spatiotemporal transfer effect: once strong 2D patch representations are in place, lightweight rotary position adaptation extrapolates them to 3D slices and video frames, letting a single 2D ViT replace modality-specific encoder stacks. Fourth, the imbalance in sequence length across modalities and the redundancy of long volumes and videos hinder scaling; our Redundancy-Guided Token Pruning (RGTP) exploits inter-slice and inter-frame redundancy to cut visual tokens by 60% while preserving performance. Across nine 2D, 3D, and video medical benchmarks, OmniV-Med achieves robust performance. Extensive component-wise ablations systematically refine the design space, yielding a simple yet efficient and scalable recipe for training high-performance unified Med-VLMs. Code is released at https://github.com/ZJUI-AI4H/Hulu-Med.</abstract>
<identifier type="citekey">jiang-etal-2026-omniv</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.275/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>6263</start>
<end>6279</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T OmniV-Med: A Scalable Training Recipe for Unified Medical Vision-Language Understanding
%A Jiang, Songtao
%A Wang, Yuan
%A Zhou, Chenyi
%A Hu, Tianxiang
%A Niu, Fucheng
%A Liu, Zuozhu
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F jiang-etal-2026-omniv
%X Practical deployment of medical vision-language models requires a single model that interprets heterogeneous clinical visual inputs: 2D images, 3D volumes, and medical videos. Yet how to train such unified models remains underexplored, as these modalities differ in spatial-temporal structure, token length, and data availability. We present OmniV-Med, a scalable training recipe built on four empirical findings. First, broad modality coverage requires balancing modality diversity with task-level supervision, motivating OmniV-Med-Instruct, a 252K corpus spanning 14 medical modalities and 11 clinical tasks. Second, we identify a modality-ordering effect: introducing 3D and video too early destabilizes optimization and weakens 2D grounding, motivating a progressive curriculum that first builds 2D alignment before extending to spatial-temporal inputs. Third, we observe an emergent spatiotemporal transfer effect: once strong 2D patch representations are in place, lightweight rotary position adaptation extrapolates them to 3D slices and video frames, letting a single 2D ViT replace modality-specific encoder stacks. Fourth, the imbalance in sequence length across modalities and the redundancy of long volumes and videos hinder scaling; our Redundancy-Guided Token Pruning (RGTP) exploits inter-slice and inter-frame redundancy to cut visual tokens by 60% while preserving performance. Across nine 2D, 3D, and video medical benchmarks, OmniV-Med achieves robust performance. Extensive component-wise ablations systematically refine the design space, yielding a simple yet efficient and scalable recipe for training high-performance unified Med-VLMs. Code is released at https://github.com/ZJUI-AI4H/Hulu-Med.
%U https://aclanthology.org/2026.emnlp-main.275/
%P 6263-6279
Markdown (Informal)
[OmniV-Med: A Scalable Training Recipe for Unified Medical Vision-Language Understanding](https://aclanthology.org/2026.emnlp-main.275/) (Jiang et al., EMNLP 2026)
ACL