ECCV 2026 Workshop On Medical Video Understanding

Welcome to the Workshop on Medical Video Understanding (MedVidU), held in conjunction with ECCV 2026.

Medical video understanding is a rapidly emerging frontier at the intersection of computer vision and healthcare. Medical procedures generate massive volumes of video data, yet the ability to automatically interpret this data — recognizing phases, detecting critical anatomical structures, grounding temporal events, and generating rich granular textual descriptions — remains an open and challenging problem. Recent advances in large multimodal models and video foundation models have created unprecedented opportunities to tackle these tasks, but standardized benchmarks that span diverse tasks and community-wide evaluation protocols are still lacking.

This workshop addresses this gap by bringing together researchers working on medical AI and video understanding. To catalyze progress, the workshop hosts the MedVidU Challenge, centered on the diverse MedVidBench dataset (CVPR 2026), spanning four domains: laparoscopic surgery, open surgery, robotic surgery, and nursing.

The MedVidU Challenge is built on the MedVidBench dataset (CVPR 2026), providing a unified evaluation across diverse tasks including Critical View of Safety (CVS) assessment, next action prediction, skill assessment, temporal action grounding, dense video captioning, and video summary & region captioning. The benchmark spans four domains: laparoscopic surgery, open surgery, robotic surgery, and nursing.

Challenge Format

The challenge runs in two phases. See the Challenge Guide for detailed step-by-step instructions.

Phase 1 — Public Leaderboard

• Download the MedVidU ECCV 2026 train/val split from UII-AI/MedVidU_ECCV2026_TrainVal and train your model.
• Run inference on the MedVidBench test set and submit your predictions to the MedVidBench Leaderboard.
• Challenge participants are required to submit a report on OpenReview (coming soon) for the MedVidU Challenge.

Phase 2 — Held-Out Evaluation

• The top 5 teams on the Phase 1 leaderboard will be further evaluated on a completely held-out test set.
• A Docker image will be released for Phase 2 submissions.

Confidentiality: We encourage teams to open-source their models, but it is not required — submitted Docker images can be kept private, are never shared or open-sourced by us, and are deleted after evaluation. See the Challenge Guide for details.

Important Dates

Registration Opens & Dataset Release (HuggingFace)	May 20, 2026	Live now
Beginning of Phase 1 (Public Validation Leaderboard)	May 20, 2026	Live now
Workshop Paper Submission (Mandatory for Challenge Participants)	July 13, 2026 (23:59 AoE)	Loading…
Phase 1 Deadline (Public Leaderboard Closes)	July 13, 2026 (23:59 AoE)	Loading…
Phase 2 Docker Image Released (Top 5 Teams)	TBA

*Deadlines are Anywhere on Earth (AoE). Timelines are subject to change.

Top 3 winners will get a chance to present their method in the workshop.

We plan to have an extended journal submission with all the phase 2 qualifiers. Prize information will be available soon.

Please submit your manuscript formatted according to the ECCV 2026 author guidelines. Accepted papers are intended to be published in the ECCV 2026 workshop proceedings.

We encourage submissions reporting novel theories, methods, and applications of video understanding in medical and surgical settings.

Note: MedVidU 2026 Challenge participants are required to submit a paper.

Topics of Interest

• Temporal action grounding in medical video
• Dense video captioning for medical procedures
• Surgical tool detection and tracking
• Medical video foundation models and multimodal large language models
• Surgical phase and action recognition
• Medical scene understanding from videos and 3D/4D reconstruction
• Privacy-preserving analysis of surgical recordings
• Explainability and trustworthiness in AI for medical videos
• Efficient video representation learning for long medical procedures

Important Dates

Paper Submission	July 13, 2026 (23:59 AoE)	Loading…
ECCV Accepted Paper Submission	July 24, 2026 (23:59 AoE)	Loading…
Paper Notification	July 31, 2026	Loading…
Camera Ready	August 15, 2026	Loading…

*All deadlines are Anywhere on Earth (AoE). Timelines are subject to change. Submission portal will be announced soon.

Half-day in-person event · Tentative schedule

Time	Activity
08:30 – 08:45	Opening Remarks
08:45 – 09:25	Keynote #1 — 30 min talk + 10 min Q&A
09:25 – 09:35	Accepted Paper Oral Presentation #1
09:35 – 09:45	Accepted Paper Oral Presentation #2
09:45 – 09:55	Accepted Paper Oral Presentation #3
09:55 – 10:35	Coffee Break / Poster Session
10:35 – 11:15	Keynote #2 — 30 min talk + 10 min Q&A
11:15 – 11:30	MedVidU Challenge: Introduction
11:30 – 12:00	Presentations from Top 3 Teams (10 min each)
12:00 – 12:15	Challenge Awards & Closing Remarks

For general inquiries about the workshop, please email medvidu@googlegroups.com.