Robotic Instrument Segmentation Sub-Challenge

Part of the Endoscopic Vision Challenge

Segmentation of robotic instruments is an important problem for robotic assisted minimially invasive surgery. It can be used for simple 2D applications such as overlay masking or 2D tracking but also for more complex 3D tasks such as pose estimation. In this challenge we invite applicants to participate in 3 different tasks: binary segmentation, multi-label segmentation and instrument recognition. Binary segmentation involves just separating the image into instruments and background, whereas multi-label segmentation requires the user to also recognize which parts of the instrument body correspond to the different articulated parts of a da Vinci robotic instrument. The final recogition task tests whether the user can recognize which segmentation corresponds to which da Vinci instrument type. 

To achieve this we are providing 8x 225-frame robotic surgical videos, captured at 2 Hz, where a trained team at Intuitive Surgical has manually labelled the different parts and types. The users are invited to test their algorithms on 8x 75-frame videos and 2x 300-frame videos which act as a test set.

The winner of this challenge will be rewarded with a $1000 prize from Intuitive Surgical.

Please note that this challenge is now over and a summary report is published on Arxiv. To use the data in your publications, please cite this work and send submissions using the same challenge submission page to obtain your results on the test dataset. You should continue to follow the same submission criteria as before.