Real Time Object Detection

This is a demonstration of object detection using the YOLOv8 (You Only Look Once) model. The out of the box model can detect 80 classes from the MS COCO dataset, and it can be trained to detect even more, as evident with the RSNA brain tumor and Baltimore aquarium datasets seen here.

The model is light enough to run on mobile CPUs and GPUs making it great for autonomous vehicles, robotics and anything else that would require real time processing.

This directory contains code, different models, outputs and evaluation metrics.

Some of my training runs got as high as 84% mAP on the Baltimore Aquarium dataset. Though the model can struggle with closely grouped instances of objects and objects that require higher resolution, we can observe that it still has passable results when used for some medical imaging datasets.