Resources#
Licensing and credits for the MedVision benchmark and the medvision_bm codebase. See the front page for the citation and the canonical project links.
Tip
For a reproducible environment, pull the published Docker image rather than resolving pinned dependencies by hand. See Installation for the setup path.
License#
The medvision_bm package is distributed under Creative Commons Attribution 4.0 (CC-BY 4.0) — see the license metadata in pyproject.toml and https://creativecommons.org/licenses/by/4.0/. The MedVision annotations themselves are released under the same CC-BY 4.0 license, which permits reuse and adaptation for academic and commercial work provided you give appropriate credit.
Warning
MedVision is a meta-dataset: it layers new annotations on top of many independently published source datasets. The CC-BY 4.0 grant covers only MedVision’s own annotations — it does not relicense the underlying imaging data. Any use of a given case must also honour the original license and usage terms of the dataset that case was drawn from. Confirming compliance for every constituent source is the user’s responsibility.
Acknowledgements#
This work was supported by UK Research and Innovation (grant EP/S02431X/1), through the UKRI Centre for Doctoral Training in Biomedical AI at the University of Edinburgh, School of Informatics.
MedVision builds directly on several open-source projects:
lmms-eval — the VLM evaluation framework underpinning the benchmark harness.
lm-evaluation-harness — LLM evaluation framework.
vLLM — high-throughput LLM/VLM inference backend.
verl — reinforcement learning for LLMs, used for RFT (via the medvision-rl fork).
TRL — supervised and preference post-training utilities.