Overview

The OphNet Phase Recognition Challenge aims to automatically identify and temporally localize surgical phases in ophthalmic surgery videos. The dataset contains 1,969 untrimmed surgical videos (~305GB in total), with 743 videos having temporal boundary annotations. For the phase recognition task, we provide 14,674 pre-trimmed phase-level video clips (~139GB) extracted from the annotated videos, where consecutive clips of the same phase have been merged to maintain semantic continuity. Phase labels appearing fewer than 15 times are mapped to ID 106 and categorized as "Others" to balance the dataset distribution. Thanks to the support from OMIA2 and MICCAI2025.
This challenge focuses exclusively on phase-level recognition task.

Dataset

For detailed information regarding OphNet dataset, please refer to: https://github.com/minghu0830/OphNet-benchmark We have released the train, val, and test datasets for model training and validation. The final ranking will be conducted on our internal dataset.

Metric

The evaluation scheme remains consistent with our setup in the APTOS2025 Big Data Challenge. The key differences are that we have a richer variety of label categories here, and the internal ranking dataset has been further expanded.

Ranking data (without GT)

coming soon.

Registration and submission portal

coming soon.

Prize

• 🏆 1st Place - TBD.
• 🥈 2nd Place - TBD.
• 🥉 3rd Place - TBD.

Timeline

15.08.2025(23:59 UTC+8) - Submission for algorithms closes.

22.08.2025(23:59 UTC+8) - Final deadline for submission of an short-paper associated with the submitted algorithm.

31.08.2025(23:59 UTC+8) - Contacting the authors of top-ranked algorithms to prepare slides for oral presentation at MICCAI.

01.09.2025 - 23.09.2025(23:59 UTC+8) - Final ranking of the results on unseen testing data.

Rules

1. No external data is allowed for model training.

2. Each team is limited to a maximum of 3 members.

OMIA Workshop-MICCAI2025 Committee

Organizers

Overview

In the OphNet phase recognition task, your goal is to develop a machine learning model capable of automatically identifying and analyzing different surgical phases in ophthalmic surgery videos. You will use the cataract surgery subset of the OphNet dataset (OphNet-Cataract), which contains 496 cataract surgery videos with detailed annotations for 35 surgical phases.
The APTOS2025 Big Data Competition has now successfully concluded!

Dataset

For detailed information regarding OphNet-Cataract dataset, please refer to: https://github.com/minghu0830/APTOS2025_OphNet

Registration and submission portal

For detailed information about the competition requirements, evaluation criteria, and rankings, please visit the official Alibaba Tianchi Competition website.

Prize

• 🏆 1st Place - US$3,000 (1 winner)
• 🥈 2nd Place - US$1,500 (1 winners)
• 🥉 3rd Place - US$500 (1 winners)

APTOS2025 Committee