Vision is an essential capability for an autonomous vehicle, as detecting and recognizing objects around it then enables the vehicle to interact with its environment. To perform this task in our submarines, we use a neural network model (YOLOv8n or YOLOv10n). This model needs to be highly reliable, lightweight and require very little training data. This is why we have chosen to train a highly specialized model for the RoboSub competition environment. Additionally, all our images are hand-labeled using Labelbox, giving us more accurate labels. (Tools and/or languages used: CUDA, Labelbox, OpenCV, Python, Pytorch, ROS, Ultralytics)