Oriented Bounding Boxes with YOLOv8

Getting started with YOLOv8

Photo by Meriç Dağlı on Unsplash

In YOLOv8.1, oriented bounding boxes (OBB) for object detection were introduced.

This means that we can now detect objects at various angles.

You can see the dramatic difference when we compare an angled image annotated with YOLOv8 OBB vs. regular object detection.

YOLOv8x vs. YOLOv8x-obb

Here’s the setup for downloading and running YOLOv8 object detection with OBB.

Choose a model:

Screenshot by author

Download the model:

from ultralytics import YOLO
model = YOLO("yolov8n-obb.pt")

Run inference:

results = model("images/image.jpg", save=True, show=True)

(Optional): I’ve noticed that resizing the image sometimes gives better results. If you want to try it out:

import torchvision.transforms as transforms
from PIL import Image

image = Image.open("images/image.jpg")
transform = transforms.Resize(1024)
resized_image = transform(image)

The object detection model is not perfect!

See, for example, what happens if we have an overview image at a slight angle.

We fail to classify the smaller ships at the back of the image.

Original photo by CHUTTERSNAP on Unsplash

Further Reading

If you want to learn more about programming and, specifically, machine learning, see the following course:

Machine Learning

Note: If you use my links to order, I’ll get a small kickback.

Links

• Oriented Bounding Boxes Object Detection