Alfred Perception: Object Detection

This repository contains code that uses the ultralyitcs package to run a Yolov8 model on a ROS node. It subscribes to a camera feed, and publishes the classes as a numpy array.

Progress

• ROS Node set up.
• Will currently be using class 39 (bottle) for integration testing. We will progress to a custom dataset with a particular set of object classes.

Future Work

• Constantly run Yolov8 and ground 2D detections to 3D map environment for remembering object locations.

Speed:

Speed: 0.6ms pre-process, 17.2ms inference, 2.7ms postprocess per image at shape (1, 3, 640, 640)