Challenge 1: Find the needle
Develop algorithms to identify the pose (position and orientation) of the metallic suture needle, with respect to the current endoscope pose.
- 3D model of needle (part of AMBF scene)
- Camera calibration
- Competitors must generate their own training data (if needed)
Each entry will be tested with different views of the surgical scene, with the needle in different locations. The entire needle will be in the field-of-view. One or two instruments may be present in the scene but will not overlap with the needle. The needle will be located between 80 mm and 200 mm from the endoscope. Lighting can vary as specified above. Stereo endoscope video will be provided at 1080p resolution at 30 fps.
Entries will be evaluated on the time to find the pose estimate and the difference between the estimated pose and the ground truth. All algorithms must output a needle pose (with respect to the ECM pose) within 60 seconds, by either publishing the pose on the specified ROS topic or calling the specified Python method. The pose difference will be determined by the distance between three fixed points on the needle: tip, middle and end. All entries that meet the time requirement will be ranked by the sum of the three distance errors. Time will be measured from when the user script is started until the estimated needle pose is received.
The evaluation script for this challenge is in the GitHub repository and can be run as follows (use
python3 as appropriate):
python evaluation.py -t <team_name> -e 1
See also the GitHub Discussions forum.
Algorithms may move the camera to better identify the needle pose. Note, however, that the simulator will add a realistic amount of error to the measurement of camera pose. Also, moving the camera will increase the time required to find the needle and will count toward the 60 second time limit. The reported needle pose should be with respect to the ECM pose when the report is published (ROS) or called (Python).