ORena SAVE FOCUS Challenge

CFP from the SDS group:  "OR-ARENA a competition designed to push the boundaries of long-context Vision-Language Models (VLMs) in surgery.

The Medical Goal

To help prevent retained surgical items by enabling models to track, contextualize, count, and localize foreign objects, such as sponges and needles.

The Technical Goal

To benchmark and advance  long-context video understanding, spanning the full spectrum from single-frame perception to procedure-level surgical understanding.

Why Participate?

• High Incentives: We have a $50,000 prize pool and aim for a joint publication of results in Nature Biomedical Engineering. The challenge takes place in parallel to the Medical Image Computing and Computer Assisted Intervention (MICCAI) conference.
• Massive VQA Dataset: Our data pool comprises over 100,000 Visual Question Answering (VQA) pairs derived from 400 laparoscopic videos (RGB). The tasks are framed as VQA problems (e.g., "How many sponges are currently in the scene?" or "Where was the needle last seen?").
• Low Entry Barrier: You do not need to tackle the entire problem at once. We offer three distinct tracks that can be completed independently. Whether you specialize in static image understanding or hour-long video reasoning, you can enter the specific track that fits your expertise; each has its own prize money pool. 
• Global Collaboration: Join a community coordinated by experts from leading institutions, including Stanford University, DKFZ (German Cancer Research Center), University of Pennsylvania, Purdue University, University College London, and Mohamed bin Zayed University of Artificial Intelligence.

Challenge Tracks

• FRAME: Basic visual perception (detection/counting in static images).
• SEGMENT: Short-term reasoning (actions/tracking in clips up to 5 min).
• PROCEDURE: The "Frontier" track (reasoning across hour-long videos).

Key Dates

• June 15: Release of second training data batch
• July 15: Launch of pre-evaluation leaderboard
• September 8: Final submission deadline
• October 1: Results announced at the MICCAI Conference (Strasbourg, France)

Whether you are a surgical AI expert or a VLM researcher looking for a high-impact, real-world application for your models, we welcome your participation. 

🌐 Website: https://or-arena.org/ 

🤗 First batch of train data on HuggingFace: https://huggingface.co/datasets/orena-dkfz/heico-focus-vqa

🛠️ Dataset utils: https://github.com/IMSY-DKFZ/orena-focus"