From YOLO to YOLOv8: Tracing the Evolution of Object Detection Algorithms

What improvements were made in the last seven years?

I have been using YOLO and its multiple versions literally every day at work for more than two years. However, if you start asking me about the specifics of YOLO — why am I using this version for that project, what’s the latest improvement, what’s going on in the world of YOLO — I’ll probably tell you it’s time for a coffee break.

Source: https://www.linkedin.com/posts/lstmeow_its-simple-you-see-the-meme-you-want-activity-7019937402848722944-qKoV?utm_source=share&utm_medium=member_desktop

This blog post is to hopefully solve that problem for all of us developers that sort of know what we are doing.

YOLO the Newborn (Release Date: June 2016)

What is YOLO? YOLO, or You Only Look Once, is an object detection model brought to us by Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi.

Why does it matter? Because of the way, the authors approached the problem. Till then, object detection was seen as a classification task of separating detected bounding boxes — looking for the best detection without spatially comparing the bounding boxes. YOLO looked at it as a regression problem and associated the probabilities of each of the detections using a single convolutional neural network (CNN). Redefining how object detection was looked at led to YOLO being faster, more accurate, and better at generalization.