YOLO (You Only Look Once) represents a family of real-time object detection algorithms in machine learning. Object detection is a core computer vision task where neural networks are employed to identify and classify objects within images. This capability powers diverse applications ranging from medical imaging analysis to autonomous vehicle systems. While multiple machine learning approaches exist for object detection, Convolusional Neural Networks (CNNs) serve as the foundational architecture for these models.

CNNs form the backbone of every YOLO iteration, enabling engineers and researchers to tackle tasks like object detection and image segmentation. The YOLO models are open-source and widely adopted across the industry, with each new version bringing improvements in accuracy, performance, and extended functionality.

Overview of YOLOv8

YOLOv8 is a cuttting-edge iteration in the YOLO series, offering state-of-the-art performance in both precision and inference speed. Although it lacks an official research paper, it follows the user-friendly design philosophy of YOLOv5. Developed by Ultralytics, YOLOv8 introduces architectural optimizations and new capabilities, making it suitable for a broad range of detection scenarios. Below are its key highlights:

• Enhanced Backbone and Neck Strutcures: Improved architectures for feature extraction and aggregation.
• Anchor-Free Segmentation Head: Utilizes an anchor-free Ultralytics head, which boosts detection efficiency and accuracy compared to traditional anchor-based methods.
• Balanced Performance: Fine-tuned to maintain an optimal trade-off between detection speed and reliability.

Ultralytics provides robust maintenance for YOLOv8, offering specialized model variants for specific computer vision tasks including detection, segmentation, image classification, and human pose estimation. The following demonstration uses the Ultralytics library to run inference.

Implementation Demo

This example uses Python and the Ultralytics library to perform inference with a YOLOv8 model. There are various ways to configure prediction settings; refer to the official documentation for details.

from ultralytics import YOLO

# Initialize a YOLOv8 small model pre-trained on the COCO dataset
detector = YOLO("yolov8s.pt")

The code above imports the library and loads the small variant of YOLOv8, which offers a strong balance between speed and accuracy. If the library is not installed, run !pip install ultralytics in your environment first. Next, we apply the model to an image.

# Run detection on an image, save outputs, and filter by confidence
predictions = detector("/content/city.jpg", save=True, conf=0.5)

This executes the detection pipeline, saves the resulting visualizations, and filters out bounding boxes with a confidence score lower than 0.5.