Getting Started with YOLOv8 on the Ultralytics Platform: Installation, Command‑Line Usage, and Model Training

1. Ultralytics Platform Overview

YOLO (You Only Look Once) is a fast, high‑accuracy object‑detection algorithm now maintained by Ultralytics, which provides an AI‑vision platform supporting detection, classification, segmentation, tracking, and pose estimation.

2. Installation and Basic Requirements

Required environment: Python ≥ 3.8 and PyTorch ≥ 1.7. Install the Ultralytics library with pip install ultralytics .

3. Command‑Line Usage

Place an image (e.g., bus.jpg ) in the working directory and run:

yolo predict model=yolov8n.pt source=bus.jpg

The command downloads the model if needed, performs inference, and saves results to runs\detect\predict .

4. Python API Usage

Import the YOLO class and run inference:

# from ultralytics import YOLO
model = YOLO("xx.pt")
results = model("bus.jpg")
print(results)

Switching tasks (detection, segmentation, pose, classification) only requires changing the model file (e.g., yolov8n-seg.pt , yolov8n-pose.pt ).

4.1 Result Structure

boxes – bounding boxes for detected objects.

masks – segmentation masks.

keypoints – pose key points.

names – class name mapping.

5. Data Annotation and Training

Use labelImg to annotate images, producing .txt files with normalized bounding‑box coordinates and a classes.txt file for class names.

Organize the dataset as:

datasets/
  game/
    images/
      train/
      val/
    labels/
      train/
      val/

Create a game.yaml configuration:

# dataset paths
train: game/images/train/
val: game/images/val/

# number of classes
nc: 1

# class names
names: ['blood']

Train via CLI:

(yolo)C:\...\yolo> yolo task=detect mode=train model=yolov8n.pt data=game.yaml epochs=300

or via Python:

from ultralytics import YOLO
model = YOLO('yolov8n.pt')
results = model.train(data='game.yaml', epochs=300)

Training outputs include best.pt , last.pt , and results.png showing loss/accuracy curves.

6. Inference with Trained Model

Run detection on a folder of images:

from ultralytics import YOLO
model = YOLO('best.pt')
results = model.predict(source='game/images/val', save=True)

Or process a video:

(yolo)C:\...\yolo> yolo task=detect mode=predict model=best.pt source="game.mp4"

7. Model Export

Export the trained model to other formats (e.g., TensorFlow.js) with three lines of code:

from ultralytics import YOLO
model = YOLO('best.pt')
model.export(format='tfjs')

8. References

Official Ultralytics website, documentation, and GitHub repository; labelImg project; and related tutorial links.