OpenSurgicalVisionOPEN SCIENCE. SHARED PROGRESS.
Sign in to CVAT
EN
OPEN SOURCE. OPEN POSSIBILITIES.

Open science.
Clearer vision.

An open-source community advancing reproducible research in surgical computer vision.

From shared datasets to transparent experiments. Built together, for everyone to explore.

RESEARCH USE ONLY – NOT FOR CLINICAL USE
Star on GitHubStart contributing
APACHE-2.0Research & education. Always open.
ANATOMICAL EXPLORATIONOSV / 001
Generated studio 3D render: a translucent amber liver revealing green bile ducts, a gallbladder and coral vascular structures, with a cyan instrument. Not patient data; not validated anatomy.
A different perspective. An open approach.
GENERATED ILLUSTRATION · NOT PATIENT DATA
FIND YOUR PATHFor engineersFor cliniciansFor contributorsFor educatorsFor researchers

Open data

Traceable sources. Clear licenses.

Reproducible models

Shared splits. Transparent methods.

Honest experiments

Negative results belong here, too.

Built by a community

Every kind of contribution counts.

THE RESEARCH WORKFLOW

Three stages. One open workflow.

Connecting anatomy, video and time – with research you can inspect.

How it works
01 / BEFORE

Reconstruct the anatomy

Explore 3D reconstruction from CT and MRI, with traceable data and annotation.

3D Slicer · MONAI
02 / DURING

Understand the scene

Research segmentation of instruments and anatomy in recorded surgical video.

SAM 2 · CVAT
03 / AFTER

Make the sequence readable

Study phases, events and reproducible timelines in retrospective recordings.

Temporal models · Evaluation

Research directions, not a clinical product. Generated concept illustrations · not patient data.

NOT JUST OPEN WEIGHTS. OPEN SCIENCE.

The result matters.
So does how we get there.

A score without context is not enough. We are building a research foundation where datasets, decisions and limitations are part of the result.

Why OpenSurgicalVision
01

Reproduce, not just report.

Fixed patient-level splits, versioned configurations and visible evaluation budgets.

02

Publish what did not work.

Negative and inconclusive experiments are research outputs, not footnotes.

03

Keep expertise in the loop.

Clinical questions require clinical review. Model suggestions are never ground truth.

FOUNDATION STAGE

Public datasets, models and experiment results will appear with verified sources. No numbers without evidence.

View roadmap

NOTES FROM THE PROJECT

Building in the open.

All perspectives
community

A place for every contribution

How to make work that is not programming visible and useful.

Read the perspective
reproducibility

Reproducibility first

Why a verifiable experiment matters more than an impressive number.

Read the perspective
negative-results

Why negative results matter

What experiments that did not improve a model are good for.

Read the perspective

Generated editorial illustrations · not experiment results.

THERE IS A PLACE FOR YOU HERE

Good research takes all of us.

Engineers, clinicians, students, writers and organizers. You do not need to write code to move research forward.

Find your first contributionMeet the roles
Are you a clinician? Your expertise belongs at the heart of this research.Explore the clinical trackStudying or teaching medicine? See real operations, train your anatomy with feedback and write research on real data.For medical schoolsStudying computer science? Solve real computer vision problems on surgical video and build an open portfolio.For CS students

START WITH THE SOURCE

Your next step is open.

Explore the repository and follow its README for the current setup instructions.

Read the quickstart
GET THE SOURCE
git clone https://github.com/opensurgicalvision/osv.git
Read. Reproduce. Contribute.