Loading...
Computer Vision

The computer vision team aims to detect where objects (such as buoys or gates) are in the water from camera data. To do this, one computer vision subteam prepared a tool that allowed MuddSub members to identify objects in images and label them with bounding boxes and categories, thus compiling a dataset. This data was then used by a second computer vision team to train and test three different neural networks — YOLOv3, MobileNet, U-net, and SqueezeDet. After implementing and examining the results of all three neural networks, the team decided to move forward with YOLOv3 since it seemed to have the highest accuracy without requiring too much computational power.

Videos